JP2009042898A - Parallel arithmetic unit and parallel operation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a parallel arithmetic unit capable of simplifying wirings among a plurality of arithmetic units. <P>SOLUTION: This parallel arithmetic unit is provided with: a plurality of units to be identified by unique unit numbers as predetermined identification numbers; and a distribution control part for outputting control data including an input value and a selection unit number as an identification number for selecting any one unit selected from among the plurality of units through a unit input bus to each of the plurality of units according as an output value to be output by any one unit among the plurality of units is input through a unit output bus. The unit executes an arithmetic operation by an arithmetic method predetermined for every unit based on the input value included in the input control data, and when the selection unit number included in the control data input to the data input part is matched with the unique unit number, the unit outputs the arithmetic result through the unit output bus to the distribution control part as an output value. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a parallel arithmetic device having a plurality of arithmetic units, and more particularly to a parallel arithmetic device that makes it possible to simplify connection between a plurality of arithmetic units.

  A neural network is known as one of parallel operations. This neural network is a mathematical model that aims to express some characteristics found in brain function. Since this neural network process can reduce the amount of information in the input data, it is a relatively small amount of computation for problems that cannot be linearly separated such as images and statistics. In many cases, a good solution can be obtained. Therefore, neural networks are applied in various fields such as pattern recognition and data mining.

  Here, if an attempt is made to realize a large-scale neural network, the amount of calculation becomes enormous and processing in a realistic time becomes difficult. As a method for solving this, (1) a method for increasing the computing power of a single processor itself, (2) a method using a parallel computing method using a plurality of processors, and (3) a function is implemented by hardware using an LSI or the like. A method, etc. can be considered. The above methods (1) and (2) are intended to deal with a huge amount of calculation by increasing the capacity of the processor, and it is possible to cope with various neural network algorithms by changing the program. it can.

  Here, in the method (1), conventionally, the computing power of the single processor itself has been increased by increasing the clock frequency of the single processor. However, in recent years, increasing the clock frequency has increased the amount of heat generation, and Moore's Law is failing because microfabrication dimensions are reaching physical limits, increasing the computing power of the single processor itself. Is getting harder. Therefore, the development of a method for increasing the computing power of a processor has shifted from the method (1) to the method (2), and it is possible to achieve both high performance by increasing computing power and suppression of heat generation. To go further, the focus is on having a larger cache and multiple compute cores. Development of this processor is also progressing so that it operates at a low clock frequency and operates at low power.

  However, in the method (2), it is difficult to efficiently operate a large number of processors, that is, to increase the degree of parallelism and to configure a network that enables huge data communication between a plurality of processors. As a general problem. Therefore, it is difficult to increase the parallel computation efficiency of a large-scale neural network by the method (2) that uses a parallel computing technique using a plurality of processors.

On the other hand, regarding the hardware implementation of the method (3), there are limitations on the neural network algorithm that can be implemented by hardware, but for specific applications, compared to the method (1) or (2), Even in the case of a low frequency, it is possible to exhibit the performance of an order of magnitude. Patent Documents 1 and 2 are known as such techniques for implementing parallel computation in hardware.
JP-A-6-195454 Japanese Patent Laid-Open No. 2006-39790

  However, in the prior arts of Patent Document 1 and Patent Document 2 relating to hardware as the method of (3), there is a wiring problem, and the circuit scale becomes enormous due to wiring, or wiring between circuits is There is a problem that it cannot be done. For example, in a multilayer neural network, it is necessary to input the output from a node with an output layer to the input of all nodes in the input layer. As the number of nodes in each layer increases, the amount of wiring rapidly increases. It will increase.

  Also, for example, in the case of an actual brain neural network, neurons are arranged and wired in a three-dimensional space, whereas in a hardware configuration such as LSI, the arrangement of components is basically two-dimensional. The wiring problem cannot be solved essentially. Even if an attempt is made to make the components three-dimensional in a laminated structure or the like, the problem of wiring remains, so the application is limited to a use where limited wiring (coupling with only the vicinity) is sufficient.

  In addition, the problem of wiring between a plurality of arithmetic units is not limited to the neural network, and it is a problem when it is necessary to input the output of the arithmetic unit to a plurality of arithmetic units at the same time. It becomes. In addition to general neural networks such as hierarchical neural networks and self-organizing maps, there are, for example, gravity many-body problems, simulation of charged many-body particles, signal filtering, and the like.

  The present invention has been made in view of such circumstances, and one of its purposes is to provide a parallel arithmetic device and a parallel arithmetic method that can simplify wiring between a plurality of arithmetic units. It is in.

  The present invention has been made to solve the above-described problem. A parallel processing device includes a plurality of units each identified by a unique unit number that is a predetermined identification number, and a plurality of units among the plurality of units. Identification for selecting any one unit selected from the input value and the plurality of units in response to an output value output from any one unit being input via the unit output bus. A distribution control unit that outputs control data including a selected unit number, which is a number, to each of the plurality of units via a unit input bus, and the unit distributes the control data via the unit input bus. A data input unit for inputting control data from the control unit; a selection unit number included in the control data input to the data input unit; Based on a unit number matching determination unit that determines whether or not the unit number matches, and an input value included in the control data input to the data input unit, the calculation is performed by a predetermined calculation method for each unit. The unit calculation unit and the unit number coincidence determination unit determine that the result of determination is the same, the calculation result that is the result of calculation by the unit calculation unit is used as the output value, and the unit output to the distribution control unit And a data output unit that outputs the data via a bus.

  According to the present invention, it is possible to simplify the connection between the plurality of arithmetic units of the parallel arithmetic device. In addition, since a plurality of units perform operations in parallel, there is an effect that operations can be performed at high speed.

  Further, according to the present invention, the distribution control unit includes a control code for selecting a calculation method of the unit calculation unit in the control data, and the unit receives the control data input to the data input unit. A control code extraction unit that extracts the control code from the unit, and the unit calculation unit executes the calculation based on the control data extracted by the control code extraction unit and the result determined by the unit number match determination unit Or a parallel control device having a unit control section for executing the calculation.

  According to the present invention, there is an effect that the distribution control unit can control all the units by using the control code. Further, when the distribution control unit controls all the units, the configuration of each unit and the configuration of the entire parallel computing device such as the connection between the distribution control unit and each unit can be simplified.

  Further, according to the present invention, the distribution control unit has a unit output value input unit that inputs an output value output from any one of the plurality of units via the unit output bus, and the unit output value. Based on the output value input to the input unit, the control data generation unit that generates the control data according to a predetermined processing procedure, and the control data generated by the control data generation unit to each of the plurality of units And a control data output unit for outputting via a unit input bus.

  According to this invention, it becomes possible for the distribution control unit to generate control data based on the output value input to the unit output value input unit according to a predetermined processing procedure, for example, a hierarchical neural network or self There is an effect that it is possible to generate control data according to a target calculation method such as an organization map.

  Further, according to the present invention, the unit includes a unit output storage unit in which the calculation result is stored, the unit calculation unit stores the calculation result in the unit output storage unit, and the data output unit includes: When the result determined by the unit number coincidence determination unit is coincident, the calculation result is read from the unit output storage unit, and the read calculation result is used as the output value as the output value to the distribution control unit. The parallel arithmetic device is characterized in that the output is made via a network.

  According to this invention, each unit stores the calculation result in the unit output storage unit, and can be output at an appropriate timing in the entire calculation processing such as a hierarchical neural network or a self-organizing map. There is an effect.

  Further, according to the present invention, the unit includes a unique unit number storage unit in which the unique unit number is stored in advance, and a selection unit number extraction unit that extracts the selected unit number from control data input to the data input unit. And an input value extraction unit that extracts the input value from the control data input to the data input unit, and the unit number match determination unit includes the selected unit number extracted by the selected unit number extraction unit, Determining whether or not the unique unit number read from the unique unit number storage unit matches, and the unit calculation unit is predetermined for each unit based on the input value extracted by the input value extracting unit. It is a parallel operation device characterized by performing an operation by the calculated operation method.

  In the present invention, in the case where the parallel arithmetic device is a parallel arithmetic device that calculates a hierarchical neural network, the unit arithmetic unit is a layer for identifying an output layer among the layers of the hierarchical neural network. An output value from the unit of the output layer identified by the layer information storage unit in which information is stored, the function storage unit in which the output function of the unit and the layer information are associated and stored in advance, and the layer information Is stored in association with a weight storage unit in which a predetermined combined load multiplied by the value and the layer information are stored in association with each other, and product sum calculation primary value information obtained by multiplying the input value from the output unit and the combined load. In response to the selection of the product-sum calculation primary value storage unit and the unit control unit, the combined load corresponding to the layer information read from the layer information storage unit is read from the weight storage unit, Multiplying the calculated combined weight by the input value input to the data input unit, and adding the multiplied value to the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit. The product-sum calculation primary value calculation unit to be stored and the product-sum calculation primary value information are read from the product-sum calculation primary value storage unit according to the selection of the unit control unit, and the layer information read from the layer information storage unit is read A corresponding output function is read from the function storage unit, an output value is calculated by substituting the read product-sum calculation primary value information into the read output function, and the calculated output value is stored in the unit output storage unit. And a unit output calculation unit to be stored.

  According to the present invention, the parallel computing device can be adapted to the hierarchical neural network.

  Further, the present invention provides a preceding unit number storage unit in which distribution unit identification information indicating which unit outputs as an output unit among the units belonging to the output layer is stored in the unit arithmetic unit. The weight storage unit includes a predetermined combined load multiplied by an output value from the unit of the output layer identified by the layer information and the distribution unit identification information, the layer information, and the layer information. Distribution unit identification information is stored in association with each other, and the product-sum calculation primary value calculation unit includes the distribution unit identification information read from the preceding unit number storage unit and the layer information read from the layer information storage unit. The corresponding connection weight is read from the weight storage unit, the read connection weight is multiplied by the input value input to the data input unit, and the multiplied value is obtained. Was added to the miracle sum calculation primary value product-sum calculating primary value information stored in the storage unit is stored, it is a parallel arithmetic apparatus according to claim.

  According to the present invention, it is possible to perform an operation of a hierarchical neural network which is composed of a plurality of layers and in which the layers are connected by a plurality of nodes.

  Further, according to the present invention, when the control code included in the control data input to the data input unit indicates that the product-sum calculation primary value is initialized, the unit control unit performs the product-sum calculation primary value. A parallel arithmetic unit that initializes product-sum calculation primary value information stored in a value storage unit to zero.

  The present invention also shows that the control code included in the control data input to the data input unit calculates and stores the product-sum calculation primary value, and the control data outputs the output value and the output value. The unit control unit stores the distribution unit identification information in the previous unit number storage unit, and based on the output value, A parallel arithmetic apparatus that performs processing of a product-sum calculation primary value calculation unit.

  In the present invention, when the control code included in the control data input to the data input unit indicates that the unit output calculation is executed, the unit control unit executes the process of the unit output calculation unit. It is a parallel arithmetic device characterized by the above.

  Further, the present invention indicates that the control code included in the control data input to the data input unit outputs an output value as an active unit, and when the control data includes a selected unit number, When the result determined by the unit number match determination unit is coincident, the control unit reads the calculation result from the unit output storage unit, and the read calculation result is used as the output value to the distribution control unit. It is a parallel operation device characterized in that it outputs via a data output unit and the unit output bus.

  Further, according to the present invention, when the control data input to the data input unit includes distribution layer information, the unit control unit displays the distribution layer information included in the control data input to the data input unit. The parallel processing device stores the information in the layer information storage unit.

  Further, the present invention includes a control code indicating that the distribution control unit initializes the product-sum calculation primary value, the output value, and the distribution unit identification information, and calculates the product-sum calculation primary value. Including a control code indicating addition storage, a control code indicating execution of the unit output calculation, a control code indicating output of an output value as the active unit, or control data including the distribution layer information The parallel processing device is characterized in that the control data is output in a predetermined order according to the number of layers of the hierarchical neural network and the number of nodes of the layer.

  Further, according to the present invention, when the parallel processing device is a parallel processing device that calculates a self-organizing map, the unit network coordinates that are unit-specific network coordinates instead of the unique unit numbers are used to identify the plurality of units. Each unit has been identified, the unit calculation unit includes a unit network coordinate storage unit in which the unit network coordinates are stored in advance, a weight storage unit in which coupling weight vectors of the units are stored, and a norm of the unit A unit norm storage unit that stores the unit norm, and control data including a control code indicating that the norm is calculated and an input vector are input from the unit control unit to the data input unit, the data input Input data vector input to the unit and the weight storage unit A norm calculation unit that calculates a norm with the extracted connection weight vector and stores the calculated norm as the unit norm in the unit norm storage unit, and a control code indicating that the norm is output and an input When control data including network coordinates is input from the unit control unit to the data input unit, the unit control unit reads the input network coordinates and the unit network coordinates read from the unit network coordinate storage unit. Is a unit that reads out the unit norm read from the unit norm storage unit via the data output unit.

  According to the present invention, there is an effect that the parallel computing device can compute the self-organizing map.

  Further, the present invention shows that the control code included in the control data input to the data input unit sets a weight, and the unit control when the input network coordinates and the input coupling weight vector are included. The unit updates the value of the input network coordinate stored in the input network coordinate storage unit to the input network coordinate included in the received control data, and the unit number match determination unit determines that the updated input network coordinate and unit It is determined whether or not the unit network coordinates read from the network coordinate storage unit match, and if the determination result of the unit number match determination unit matches, the unit control unit is stored in the weight storage unit Updating the existing connection weight vector to the input connection weight vector included in the received control data. A parallel operation and wherein.

  Further, the present invention indicates that the control code included in the control data input to the data input unit sets the learning speed and the proximity distance, and includes the input learning speed and the input proximity distance. The unit control unit updates the learning speed stored in the learning speed storage unit to the input learning speed included in the received control data, and converts the reference distance stored in the distance storage unit into the control data received. The parallel computing device is characterized by updating to the included input neighborhood distance.

  Further, the present invention shows that the control code included in the control data input to the data input unit sets the number of dimensions of the real vector, and when the input dimension number N is included, the unit control unit The parallel arithmetic device is characterized in that the number of dimensions stored in the real vector dimension number storage unit is updated to the number of input dimensions included in the received control data.

  Further, the present invention indicates that the control code included in the control data input to the data input unit calculates a norm, and when the input code includes an input vector, the unit control unit includes an input data storage unit The norm calculation unit calculates a norm between the updated input vector and the combined weight vector read from the weight storage unit, and the calculated norm is a unit norm. Is stored in the unit norm storage unit.

  Further, the present invention indicates that the control code included in the control data input to the data input unit outputs a norm, and if the control code includes input network coordinates, the unit control unit stores the input network coordinate The value of the input network coordinates stored in the unit is updated to the input network coordinates included in the received control data, and the unit number match determination unit reads out the updated input network coordinates and the unit network coordinate storage unit. It is determined whether or not the unit network coordinates match, and if the determination result of the unit number match determination unit matches, the unit control unit outputs the unit norm read from the unit norm storage unit as the data output Output to the distribution control unit via the unit and the unit output bus. A parallel operation apparatus according to symptoms.

  Further, the present invention indicates that the control code included in the control data input to the data input unit is learned, and when the input network coordinate is included, the unit control unit stores the input network coordinate storage The input network coordinates stored in the unit are updated to the input network coordinates included in the control data, and the arithmetic unit learns the control code included in the control data input to the data input unit. , The distance between the unit network coordinates read from the unit network coordinate storage unit and the input network coordinates read from the input network coordinate storage unit is calculated, and the calculated distance is the distance storage unit By determining whether the distance is less than or equal to the distance read from the unit network coordinates When the unit network coordinates read from the unit are in the vicinity of the input network coordinates read out from the input network coordinate storage unit, and the determination result of the proximity determination unit is near A value obtained by calculating the difference between the input data vector read from the input data storage unit and the combined weight vector read from the weight storage unit, and multiplying the calculated difference by the learning speed coefficient read from the learning speed storage unit Is added to the connection weight vector stored in the weight storage unit, and the weight update unit updates the connection weight vector stored in the weight storage unit.

  Further, the present invention indicates that the control code included in the control data input to the data input unit outputs a learning result, and when the input network coordinates are included, the unit control unit The value of the input network coordinate stored in the network coordinate storage unit is updated to the input network coordinate included in the received control data, and the unit number coincidence determination unit stores the updated input network coordinate and the unit network coordinate It is determined whether or not the unit network coordinates read from the unit match, and if the determination results of the unit number match determination unit match, the unit control unit reads the combined weight vector read from the weight storage unit Is output to the distribution control unit via the data output unit and the unit output bus. A parallel operation and wherein.

  Also, the present invention provides a control code for setting the weight, a control code for setting the learning speed and a neighborhood distance, and a control for setting the number of dimensions of a real vector. A plurality of control data including a code, a control code indicating calculating a norm, a control code indicating outputting a norm, a control code indicating learning, or a control code indicating outputting a learning result The parallel processing device is characterized in that it outputs in a predetermined order according to the number of units and the number of learnings.

  Further, according to the present invention, the unit output bus and the unit output bus that connect the plurality of units and the distribution control unit are separated by a first connection unit and a second connection unit that can be connected to each other, A parallel arithmetic device characterized in that the first device includes the plurality of units and the first connection unit, and the second device includes the distribution control unit and the second connection unit. is there.

  In addition, according to the present invention, a plurality of units each identified by a unique unit number, which is a predetermined identification number, and an output value output from any one of the plurality of units are transmitted to the unit output bus. Control data including the inputted input value and a selection unit number which is an identification number for selecting any one unit selected from the plurality of units in response to being inputted through the plurality of units. And a distribution control unit that outputs to each of the units via a unit input bus, wherein the unit is controlled by the distribution control unit via the unit input bus. When data is input, the selected unit number included in the control data input to the data input unit matches the unique unit number. Whether or not is calculated by a predetermined calculation method for each unit based on the input value included in the input control data. If the determined result is coincident, the calculation is performed. The parallel calculation method is characterized in that a calculation result as a result is output as the output value to the distribution control unit via the unit output bus.

<First basic configuration>
Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a schematic block diagram showing a configuration (first basic configuration) of a parallel arithmetic device according to an embodiment of the present invention. Here, a case where the number of units U1 to U10 is ten will be described. Details of this unit will be described later.

  The plurality of units U1 to U10 are daisy chained via the daisy chain control bus 1 in a daisy chain order that is a predetermined order. Here, this daisy chain order will be described assuming that the order of unit U1, unit U2, unit U3,..., Unit U10 is predetermined. The daisy chain control bus 1 connects adjacent units in the daisy chain order. For example, the daisy chain control bus 1 connects the unit U1 and the unit U2, the unit U2 and the unit U3,..., The unit U10 and the unit U1. The daisy chain control bus 1 connects units U1 to U10 in a daisy chain sequence. Therefore, the unit U10 and the unit U1 are connected via the daisy chain control bus 1.

  Each of the units U1 to U10 transmits and receives tokens between the units via the daisy chain control bus 1 in the daisy chain order. This token is a signal indicating that any one of the daisy chained units U1 to U10 has a unit, and that the unit having this token is an output unit. As for the unit, only a unit having a token becomes an output unit, and all units become input units.

  Each unit performs an operation by a predetermined operation method for each unit based on an input value from the amplifier 4 input via the unit output bus 2. Also, each unit receives a token from the previous unit via the daisy chain control bus 1 in the daisy chain order, and passes the token to the next unit via the daisy chain control bus 1 in the daisy chain order. At the same time, the calculated result is output as an output value to the amplifier 4 via the unit output bus 2.

  The amplifier (repeater) 4 receives an output value output from any one of the plurality of units U1 to U10 via the unit output bus 2, and inputs the input output value to each of the plurality of units U1 to U10. And output as an input value via the unit input bus 3. The amplifier 4 electrically amplifies the input output value signal and outputs it as an input value.

  One end of the unit output bus 2 is connected to the output terminal P1 of the parallel arithmetic device. Further, the input terminal of the amplifier 4 is connected to the input terminal P2 of the parallel arithmetic device. The unit U1 is connected to the trigger input terminal P3 through the trigger input line 5. The output terminal P1, the input terminal P2, and the trigger input terminal P3 are connected to a control device outside the parallel arithmetic device. The control device inputs data to the parallel arithmetic device via the output terminal P1, the input terminal P2, and the trigger input terminal P3, acquires the arithmetic result from the parallel arithmetic device, and controls the parallel arithmetic device.

  Next, the configuration of each unit will be described with reference to FIG. Since units U1 to U10 have the same configuration, only the configuration of unit U2 will be described here. The unit U2 includes a token input unit 12, a token output unit 13, a data input unit 11, a data output unit 14, a unit calculation unit 15, and a unit output storage unit 16.

  The token input unit 12 inputs a token from the unit in the previous order in the daisy chain via the daisy chain control bus 1. For example, the token input unit 12 of the unit U2 inputs a token from the unit U1 which is the unit in the previous order in the daisy chain via the daisy chain control bus 1. Note that the token input unit 12 of the unit U1 not only receives a token from the unit U10 via the token daisy chain control bus 1, but also receives a token from the control device via the trigger input terminal P3 and the trigger input line 5. Is input via.

  The token output unit 13 outputs the token via the daisy chain control bus 1 to the next unit in the order in which the tokens are daisy chained in response to the token input to the token input unit. For example, the token output unit 13 of the unit U2 outputs a token to the unit U3 which is the unit that is the next unit in the daisy chain order via the daisy chain control bus 1.

The data input unit 11 inputs an input value from the amplifier 4 via the unit output bus 2. The unit output storage unit 16 stores a calculation result that is a result calculated by the unit calculation unit 15. The unit calculation unit 15 calculates based on the input value input to the data input unit 11 by a calculation method predetermined for each unit. Further, the unit calculation unit 15 causes the unit output storage unit 16 to store a calculation result that is a calculation result.

  The data output unit 14 outputs, as an output value, a calculation result that is a result of calculation by the unit calculation unit 15 to the amplifier 4 via the unit output bus 2 in response to the token input to the token input unit 12. . Further, the data output unit 14 reads out the calculation result from the unit output storage unit 16 in response to the token input to the token input unit 12, and uses the read calculation result as an output value to the amplifier 4 for the unit output bus 2. Output via.

  As will be described later, the parallel arithmetic device inputs a token that is a trigger to the unit U1 via the trigger input terminal P3, and the tokens are sequentially passed between the units in order from the unit U1, thereby the parallel arithmetic device. The processing in each unit is executed. The operation of the parallel arithmetic device will be described by taking as an example a case where the parallel arithmetic device described below is applied to a hierarchical neural network.

<First Embodiment>
<Unit configuration when applied to hierarchical neural network>
Next, a configuration when the parallel arithmetic device is applied to a hierarchical neural network will be described. Also in the hierarchical neural network, the configuration of the entire parallel arithmetic device is the same as that of the parallel arithmetic device in FIG. 1, and only the configuration of the unit is different. Therefore, the configuration of the unit when applied to the hierarchical neural network will be described with reference to FIG.

  The unit includes a data input unit 101, a token input unit 102, a token output unit 103, and a data output unit 104. The unit includes a weight storage unit 130, a layer information storage unit 133, a unit number storage unit 134, a product-sum calculation primary value storage unit 131, a function storage unit 132, a unit output storage unit 135, and a data output. A flag storage unit 136 and a unit identification information storage unit 137 are provided. The unit includes a counter 114, a product-sum calculation primary value calculation unit 111, a layer information calculation unit 113, and a unit output calculation unit 112.

  Here, the data input unit 101, token input unit 102, token output unit 103, and data output unit 104 of FIG. 3 are connected to the data input unit 11, token input unit 12, token output unit 13, and data output unit 14 of FIG. , Respectively. Further, the unit output storage unit 135 in FIG. 3 corresponds to the unit output storage unit 16 in FIG. Descriptions of common functions in the configurations corresponding to those in FIGS. 2 and 3 are omitted.

  Also, the weight storage unit 130, the layer information storage unit 133, the unit number storage unit 134, the product-sum calculation primary value storage unit 131, the function storage unit 132, the data output flag storage unit 136, the unit of FIG. The identification information storage unit 137, the counter 114, the product-sum calculation primary value calculation unit 111, the layer information calculation unit 113, and the unit output calculation unit 112 correspond to the unit calculation unit 15 in FIG.

  The layer information storage unit 133 stores layer information for identifying an active layer (output layer) among the layers of the hierarchical neural network. The unit identification information storage unit 137 stores unit identification information indicating which unit of the units belonging to the active layer is outputting. In the function storage unit 132, the output function of the unit and the layer information are stored in advance in association with each other. The output function is a predetermined function such as a sigmoid function, a step function, or a piecewise linear function.

  In the weight storage unit 130, a predetermined coupling load multiplied by the output value from the unit of the active layer identified by the layer information is stored in association with the layer information. Further, the weight storage unit 130 stores a predetermined coupling load multiplied by an output value from the unit of the layer identified by the layer information and the unit identification information, and the layer information and the unit identification information in association with each other. ing.

  In the unit number storage unit 134, the layer information and the number of units of the hierarchical neural network belonging to the layer identified by the layer information are stored in advance in association with each other. The product-sum calculation primary value storage unit 131 stores product-sum calculation primary value information obtained by multiplying the input value from the unit corresponding to the preceding unit and the coupling load. The data output flag storage unit 136 stores a data output flag indicating whether or not the data output unit 104 has already output an output value. The counter 114 increments the value of the unit identification information stored in the unit identification information storage unit 137 every time an input value is input to the data input unit 101.

  The unit number comparison unit 115 reads layer information from the layer information storage unit 133, reads the number of units corresponding to the read layer information from the unit number storage unit 134, and reads and reads unit identification information from the unit identification information storage unit 137. The unit identification information value is compared with the number of read units, and if the comparison results match, a comparison match signal indicating that the read unit identification information value matches the read unit number is output. . Further, the unit number comparison unit 115 compares the read unit identification information value with the read unit number, and if the comparison result matches, the unit identification information stored in the unit identification information storage unit 137 Reset the value.

  The product-sum calculation primary value calculation unit 111 reads the connection load corresponding to the layer information read from the layer information storage unit 133 from the weight storage unit 130, the read connection load, and the input value input to the data input unit 101 , And the multiplied value is added to and stored in the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131.

  Further, the product-sum calculation primary value calculation unit 111 reads out and reads the combined load corresponding to the unit identification information read from the unit identification information storage unit 137 and the layer information read from the layer information storage unit 133 from the weight storage unit 130. The combined weight and the input value input to the data input unit 101 are multiplied, and the multiplied value is added to and stored in the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131.

  The unit output calculation unit 112 reads the product-sum calculation primary value information from the product-sum calculation primary value storage unit 131, reads the output function corresponding to the layer information read from the layer information storage unit 133 from the function storage unit 132, and reads The output value is calculated by substituting the read product-sum calculation primary value information into the output function, and the calculated output value is stored in the unit output storage unit 135.

Further, the unit output calculation unit 112 reads the product sum calculation primary value information from the product sum calculation primary value storage unit 131 in response to the input of the comparison coincidence signal from the unit number comparison unit 115, and the layer information storage unit 133 The output function corresponding to the layer information read out from is read from the function storage unit 132, the output value is calculated by substituting the read product-sum calculation primary value information into the read output function, and the calculated output value is stored in the unit output Stored in the unit 135.
The unit output calculation unit 112 reads the product-sum calculation primary value information from the product-sum calculation primary value storage unit 131, and then stores the value of the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131. Is reset to 0.

  The layer information calculation unit 113 increments the value of the layer information stored in the layer information storage unit 133 in response to the comparison match signal input from the unit number comparison unit 115. The data output unit 104 sets the data output flag of the data output flag storage unit 136 as output in response to the token input to the token input unit 102.

  The token output unit 103 reads the data output flag from the data output flag storage unit 136 in response to the input value being input to the data input unit 101, and the read data output flag is set to have been output. Output a token (token output procedure). In addition, the token output unit 103 outputs a token and sets the data output flag of the data output flag storage unit 136 to non-output.

  The token output unit 103 reads the data output flag from the data output flag storage unit 136 in response to the input value being input to the data input unit 101, and the read data output flag is set to have been output. In this case, instead of the token output procedure of outputting a token, the data output flag storage unit 136 determines that the input value from the data input unit 101 is input to the product-sum calculation primary value calculation unit 111. If the data output flag is read and the read data output flag is set to be output, a token may be output.

<Unit operation when applied to hierarchical neural network>
Next, the operation of the unit when applied to the hierarchical neural network of FIG. 3 will be described using FIG. 4 and FIG. First, the operation when an input value is input to the unit will be described with reference to FIG.

  First, an input value is input from the amplifier 4 to the data input unit 101 via the unit input bus 3 (step S100). Next, in step S100, in response to the input value being input to the data input unit 101, the counter 114 increments the value of the unit identification information stored in the unit identification information storage unit 137 by one (step S100). S101).

  In addition, in step S100, in response to the input value being input to the data input unit 101, the product-sum calculation primary value calculation unit 111 reads the unit identification information and the layer information storage unit 133 read from the unit identification information storage unit 137. The combined weight corresponding to the layer information read out from is read out from the weight storage unit 130, the read out combined weight is multiplied by the input value input into the data input unit 101, and the multiplied value is stored as a product-sum calculation primary value storage. The product-addition calculation primary value information stored in the unit 131 is added and stored.

  Next, the unit number comparison unit 115 reads the layer information from the layer information storage unit 133, reads the number of units corresponding to the read layer information from the unit number storage unit 134, and receives the unit identification information from the unit identification information storage unit 137. A comparison match signal indicating that the value of the read unit identification information matches the number of read units when the read and read unit identification information values are compared with the number of read units and the comparison results match. Is output (step S103). Further, the unit number comparison unit 115 compares the read unit identification information value with the read unit number, and the unit identification information stored in the unit identification information storage unit 137 when the comparison result matches. Reset the value of.

  Next, when the comparison results in step S103 match, the layer information calculation unit 113 stores the comparison match signal from the unit number comparison unit 115 and stores it in the layer information storage unit 133. The layer information value is incremented (step S104).

  Next, when the comparison results in step S103 match, the unit output calculation unit 112 responds to the input of the comparison match signal from the unit number comparison unit 115, and the product-sum calculation primary value storage unit 131. The product sum calculation primary value information is read out, the output function corresponding to the layer information read out from the layer information storage unit 133 is read out from the function storage unit 132, and the read product sum calculation primary value information is substituted into the read out output function. Then, the output value is calculated, and the calculated output value is stored in the unit output storage unit 135 (step S105). Next, the unit output calculation unit 112 reads the product-sum calculation primary value information from the product-sum calculation primary value storage unit 131, and then stores the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131. The value of is reset to 0.

  Next, or when the result of comparison in step S103 is inconsistent, the token output unit 103 determines that the data output flag is in accordance with the input value input to the data input unit 101 in step S100. A data output flag is read from the storage unit 136, and it is detected whether or not the read data output flag has been output (step S106). If the data output flag is set to output, a token is output. (Step S107).

  In addition, the token output unit 103 outputs a token in step S107, sets the data output flag of the data output flag storage unit 136 to non-output (step S108), and ends the process. On the other hand, when the data output flag is set to non-output in step S106, the token output unit 103 ends the process.

  Next, the operation when a token is input to the unit will be described with reference to FIG. First, tokens from the unit in the previous order in the daisy chain via the daisy chain control bus 1 are input to the token input unit 102 (step S200).

  Next, the data output unit 104 reads the output value (calculation result) from the unit output storage unit 135 in response to the token input to the token input unit 102, and outputs the read output value (calculation result) as the output value. Is output to the amplifier 4 via the unit output bus 2 (step S201). Next, the data output unit 104 sets the data output flag of the data output flag storage unit 136 to “already output” in response to the token input to the token input unit 102 (step S202), and ends the process. .

<Overall configuration of hierarchical neural network>
Next, the overall configuration of the hierarchical neural network according to the present embodiment will be described.
Here, as shown in FIG. 6, a case where there are three layers of an input layer L1, a hidden layer L2, and an output layer L3 as a hierarchical neural network will be described. Each layer is identified by the layer information, and the layer information of the layers is identified such that the layer information increases in order of the layers. For example, the layer information of the input layer L1 is identified as 1, the layer information of the hidden layer L2 is 2, and the layer information of the output layer L3 is identified as 3.

  The input layer L1 has three nodes P101, P102, and P103, and the hidden layer L2 has ten nodes P201, P202, P203, P204, P205, P206, P207, P208, P209, and P210. A case where there are nodes and the output layer L3 has three nodes P301, P302, and P303 will be described. The output of each node of the input layer L1 is input to the input of each node of the hidden layer L2, and the output of each node of the hidden layer L2 is input to the input of each node of the output layer L3.

  In each layer, each node is identified by unit identification information, and the unit identification information of each node is set so that the unit identification information is increased by one in the predetermined node arrangement order. For example, P101 has unit identification information of 1, P102 has unit identification information of 2, and P103 has unit identification information of 3.

In the present embodiment, each unit functions as a node of the hidden layer L2 and the output layer L3. In this example, the units U1 to U3 function as nodes of the hidden layer L2 and the output layer L3, and the units U4 to U10 function as nodes of the hidden layer L2. Here, the control device sequentially outputs the output value as the input layer L1 to each unit as the hidden layer L2 via the amplifier 4 and the unit output bus 2. Therefore, here, each unit functions as a hidden layer L2 and an output layer L3.
For example, the unit U1 functions as a node P201 in the hidden layer L2 and a node P301 in the output layer L3. The unit U3 functions as a node P203 in the hidden layer L2 and a node P303 in the output layer L3. The unit U4 functions as the node P204 of the hidden layer L2.

  Here, when a value is input to a certain unit, the input from the input layer L1, the hidden layer L2, or the output layer L3 is stored in the layer information storage unit 133 of each unit. It is determined based on certain layer information.

<Overall operation of hierarchical neural network>
Next, an overall operation as an example of the hierarchical neural network according to the present embodiment will be described. Here, a case will be described in which each unit functions as a node of the output layer L3, and any one unit to which a token is input functions as an active node of the hidden layer L2. In this case, since each unit functions as a node of the output layer L3 and the active layer is the hidden layer L2, 2 is stored as the value of the layer information in the layer information storage unit 133 of each unit.

  First, the unit U1 which is the node P201 reads the output value from the unit output storage unit 135 and outputs the output value when the token is input from the control device, and sets the data output flag of the data output flag storage unit 136. Set as output completed (procedure A100). Next, the output value from the node P201 is input to the data input unit 101 of each unit of the units U1 to U10 via the unit output bus 2, the amplifier 4, and the unit input bus 3.

Next, each of the units U1 to U10 has unit identification information and a layer information storage unit 133 read from the unit identification information storage unit 137 by the product-sum calculation primary value calculation unit 111 based on the input value from the unit U1. The combined weight corresponding to the layer information read out from is read out from the weight storage unit 130, the read out combined weight is multiplied by the input value input into the data input unit 101, and the multiplied value is stored as a product-sum calculation primary value storage. The product-addition calculation primary value information stored in the unit 131 is added and stored (procedure A101).
Here, the product-sum calculation primary value calculation unit 111 of each unit uses the weight storage unit 130 to determine the combined load corresponding to the unit identification information read from the unit identification information storage unit 137 and the layer information read from the layer information storage unit 133. By reading from, each unit functions as a node of the output layer L3 based on the layer information, and identifies the layer to which the active unit belongs as the hidden layer L2 based on the layer information. Based on the identification information, each node in the active hidden layer L2 is identified, and the coupling load corresponding to the active node in the hidden layer L2 is set as an output value from the node in the active hidden layer L2. It is possible to multiply.

  Here, since the data output flag of the data output flag storage unit 136 of the unit U1 is set as already output, the token output unit 103 of the unit U1 responds to the input value being input to the data input unit 101. Then, the token is output and the data output flag of the data output flag storage unit 136 is set as not output (procedure A102).

  Next, the token output from the unit U1 is input via the daisy chain control bus 1 to the token input unit 12 of the unit U2 which is the next unit in the daisy chain order (procedure A103). .

  In response to the input of the token to the token input unit 12, the unit U2 outputs the unit U1 as the node P101 in the procedure A100 described above from the unit output storage unit 135 in the same manner as when the token is input. The value is read, the output value is output, and the data output flag in the data output flag storage unit 136 is set as already output (procedure A104).

  Thereafter, similarly, the procedure A101 to the procedure A103 are repeated, and the unit U10 that is the node P210 of the hidden layer L2 outputs the output value and sets the data output flag of the data output flag storage unit 136 as already output.

  Next, the units U1 to U3 corresponding to the nodes P301 to P303 of the output layer L3 are similar to the procedure A101, based on the input value from the unit U10, the product-sum calculation primary value calculation unit 111 of each unit. However, the combined load corresponding to the unit identification information read from the unit identification information storage unit 137 and the layer information read from the layer information storage unit 133 is read from the weight storage unit 130, and the read combined load and the data input unit 101 are The inputted input value is multiplied, and the multiplied value is added to and stored in the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131 (procedure A105).

Here, in response to the input value from the unit U10 being input to the data input unit 101 of each unit, the counter 114 of each unit displays the value of the unit identification information stored in the unit identification information storage unit 137. And the value of the unit identification information becomes 3 (procedure A106).
The unit number storage unit 134 of each unit stores in advance the layer information and the number of units of the hierarchical neural network belonging to the layer identified by the layer information in association with each other. For example, in the case of the hierarchical neural network of FIG. 6, the value 1 which is the layer information of the input layer L1 and the number of units of the hierarchical neural network, that is, the number of nodes of the input layer L1 are associated with each other. Stored in advance. Further, for example, in the case of the hierarchical neural network of FIG. 6, the value 2 that is the layer information of the hidden layer L2 and the number of units of the hierarchical neural network, that is, the number of nodes of the hidden layer L2 are 10 Pre-stored in association.

  Next, the unit number comparison unit 115 of each unit reads the layer information from the layer information storage unit 133, reads the number of units corresponding to the read layer information from the unit number storage unit 134, and reads the unit number from the unit identification information storage unit 137. The identification information is read out, the value of the read unit identification information is compared with the number of read units, and the comparison result matches, so that the value of the read unit identification information matches the number of units read. A comparison coincidence signal is output, and the value of the unit identification information stored in the unit identification information storage unit 137 is reset (procedure A107).

  Next, the unit output calculation unit 112 of each unit reads the product sum calculation primary value information from the product sum calculation primary value storage unit 131 in response to the input of the comparison coincidence signal from the unit number comparison unit 115, and An output function corresponding to the layer information read from the information storage unit 133 is read from the function storage unit 132, an output value is calculated by substituting the read product-sum calculation primary value information into the read output function, and the calculated output value Is stored in the unit output storage unit 135 (procedure A108).

  Here, the unit output calculation unit 112 of each unit reads the output function corresponding to the layer information read from the layer information storage unit 133 from the function storage unit 132, and therefore, by the output function for each unit corresponding to the output layer L3, The output value of the unit can be calculated. In addition, since the output value calculated as the node of the output layer L3 is stored in the unit output storage unit 135 of each unit, when each unit receives a token, each unit becomes an active node of the output layer L3. An output value can be output.

  Further, the layer information calculation unit 113 of each unit increments the value of the layer information stored in the layer information storage unit 133 in response to the input of the comparison coincidence signal from the unit number comparison unit 115, and The information value is 3. For this purpose, each unit outputs an output value as an active node of the output layer L3 after that any one unit to which a token is input.

  In the above, a token is output from the unit U3 to the unit U4 in the output layer L3, and then the unit U4 to which the token is input outputs an output value as a node of the output layer L3. Since there is no node corresponding to the unit U4 as a node of the output layer L3 in the network, it is desirable that the unit U4 does not output an output value as a node of the output layer L3. Alternatively, the unit U4 may output the output value as 0 as a node of the output layer L3. Alternatively, in each unit, the connection load multiplied by the output value from the unit U4 as the output layer L3 may be set to 0 and stored in the weight storage unit 130 in advance.

  Thus, the unit U4 output value or the connection load of each unit that receives the output value from the unit U4 is set in advance in accordance with the number of nodes in each layer of the hierarchical neural network. On the other hand, even when there is no node corresponding to the unit in a certain layer, the parallel arithmetic device of this embodiment can operate as a hierarchical neural network without any problem.

As described above, when this embodiment is applied to a hierarchical neural network, the number of neurons that fire at one time is limited to one, that is, the node that outputs an output value in the output layer, By limiting to one unit to which a token is input, an effect of reducing an increase in wiring between units can be obtained.
Further, since the output value from the unit which is a node of one output layer is distributed and inputted to the unit which is a node which becomes the input layer at a time via the unit input bus 3, it is necessary for the distribution. It has the effect of reducing time. In addition, since a plurality of units perform in-unit operations in parallel, the overall operation is fast.

  For example, when calculating a hierarchical neural network, if a conventional processor is used, the calculation time is approximately calculated as calculation time = (number of synapses × number of steps required for one calculation ÷ clock frequency). On the other hand, when the parallel arithmetic device according to the present embodiment is used, calculation time = (number of neurons × one neuron firing cycle). Here, since the number of synapses is proportional to the square of the number of neurons, the calculation time of the hierarchical neural network using the parallel arithmetic device according to this embodiment is calculated as compared with the case of using a conventional processor. Time can be greatly reduced.

  In addition, since each unit changes the layer of the hierarchical neural network and functions as a node on the input side and a node on the output side, it is possible to perform multi-layer node operations with a small number of units. There is a possible effect.

  In the first embodiment described above, the control device sequentially outputs output values as the input layer L1, and therefore, each unit functions as a node of the hidden layer L2 and the output layer L3. Each unit may function as a node of the input layer L1. For example, the control device sequentially outputs output values corresponding to the nodes of the input layer L1, and the unit sequentially stores output values from the control device in the unit output storage unit 135 as nodes of the input layer L1. Next, the unit to which the token is input outputs an output value as an active unit of the input layer L1. In this way, each unit can also function as a unit of the input layer L1, the hidden layer L2, and the output layer L3.

  In the first embodiment, the case where there are three layers of the input layer L1, the hidden layer L2, and the output layer L3 as the hierarchical neural network has been described. However, in the present embodiment, the hierarchical neural network The hierarchy is not limited to three layers, and can be applied to any hierarchy.

<Second Embodiment>
<Unit structure when adapted to the self-organizing map>
Next, the case where the parallel arithmetic device of this embodiment is applied to a self-organizing map will be described. The self-organizing map learns the statistical properties of multivariate data by performing the following processes, arranges similar data in close proximity, and visualizes the data: It is possible.

  Each unit has a unit connection weight vector. First, an input data vector is input to each unit, and each unit calculates a norm that is the difference between the input data vector and the connection weight vector (procedure A200). Next, the output unit with the smallest norm is selected as the winner unit (procedure A201). Next, learning is performed by changing the value of the coupling weight vector with a predetermined function so that the selected winner unit and its neighboring units are closer to the input data vector (procedure A203). This procedure A200 to procedure A203 are repeated.

  When the parallel computing device of this embodiment is applied to a self-organizing map, the data input to each unit is different in the above-described procedure A200, procedure A201, and procedure A202. The input data is identified by combining identification information for identifying the procedure A200, the procedure A201, and the procedure A202.

  As shown in FIG. 9A, the input value from the unit input bus 3 has a norm part and a data part. Further, as shown in FIG. 9B, when the value of the norm part is −2, the norm part of the input value is an identifier indicating that “the value of the data part is an input vector”, and the data part Is the input vector. As shown in FIG. 9C, when the value of the norm part is −1, the norm part is “neighboring determination identification information”, and the data part is network coordinates. Further, as shown in FIG. 9D, when the value of the norm part is zero or a positive value, the value of the norm part is the norm, and the data part is the network coordinates. Details of these will be described later.

  Next, the configuration of the units of the parallel arithmetic device when applied to the self-organizing map will be described using FIG. 7 and FIG. Here, each of the plurality of units is identified in advance by coordinates identified by network coordinates. Since each unit has the same configuration as in the first embodiment, only the configuration of one unit will be described here.

  7 or 8, the data input unit 201 corresponds to the data input unit 11, the data output unit 204 corresponds to the data output unit 14, and the token input unit 202 corresponds to the token input unit 12. The token output unit 203 corresponds to the token output unit 13. The unit norm storage unit 236, the input norm storage unit 237, the unit network coordinate storage unit 234, and the input network coordinate storage unit 233 correspond to the unit output storage unit 16. The other configuration corresponds to the unit calculation unit 15.

  In the unit network coordinate storage unit 234, unit network coordinates, which are unit network coordinates, are stored in advance. The input network coordinate storage unit 233 stores input network coordinates that are input network coordinates. The weight storage unit 231 stores unit connection weight vectors in advance. The input data storage unit 230 stores input input data vectors. The unit norm storage unit 236 stores the unit norm. The input norm storage unit 237 stores the input norm.

  The learning speed storage unit 235 stores a learning speed coefficient that is a coefficient for determining the learning speed in advance. In addition, the learning speed storage unit 235 stores a learning step number indicating the number of times of learning and the learning speed coefficient in association with each other in advance. The distance storage unit 232 stores in advance a reference distance that is a reference for determining whether or not the distance between the network coordinates is near. The distance storage unit 232 stores the number of learning steps and the reference distance in association with each other.

The data input determination unit 222 determines whether or not the input value input to the data input unit 201 is an input data vector. If the input value is an input data vector, the input data that is an input value to the data extraction unit 210 is determined. When the vector is output and not the input data vector, the input value is output to the network coordinate extraction unit 211 and the norm extraction unit 212.
The data input determination unit 222 is input to the data input unit 201 by determining whether the norm value of the input value is −2 that is a value indicating that the input value is an input data vector. It is determined whether the input value is an input data vector.

  If the input value input to the data input unit 201 is an input data vector, the data extraction unit 210 causes the input data storage unit 230 to store the input data vector input. Further, the data extraction unit 210 outputs the input data vector that has been input to the norm calculation unit 214. In the data extraction unit 210, the counter 213 outputs a count-up signal each time an input data vector is input.

  Each time an input data vector is input to the unit, the counter 213 counts up the number of steps that is the number of times the input data vector is input to the unit. The counter 213 counts up the number of steps each time a count up signal is input from the data extraction unit 210.

  When the input value input to the data input unit 201 is an input data vector, the norm calculation unit 214 calculates a norm between the input data vector input and the combined weight vector read from the weight storage unit 231. The calculated norm is stored in the unit norm storage unit 236 as a unit norm.

  When the input value input to the data input unit 201 is a combination of network coordinates and norm, the input norm unit 250 stores the input network coordinates as input network coordinates in the input network coordinate storage unit 233. At the same time, the input norm is stored in the input norm storage unit 237 as an input norm. The input norm unit 250 includes a network coordinate extraction unit 211, a norm extraction unit 212, and an input norm determination unit 218.

  The network coordinate extraction unit 211 extracts network coordinates from the input values that have been input, and stores the extracted network coordinates in the input network coordinate storage unit 233 as input network coordinates. The norm extraction unit 212 extracts a norm from the input value that has been input, and outputs the extracted norm to the input norm determination unit 218.

The input norm determination unit 218 determines whether the input norm is a norm based on the norm value input from the norm extraction unit 212 or is identification information indicating identification information indicating that the network coordinates and the proximity determination are executed. Whether it is identification information or not is determined. If it is a norm, the input norm is stored in the input norm storage unit 237 as an input norm. Conversely, when the input norm is the proximity determination identification information, the input norm determination unit 218 outputs the proximity determination identification information to the proximity determination unit 215.
For example, when the norm value input from the norm extraction unit 212 is −1, the input norm determination unit 218 determines that the input norm is proximity determination identification information, and the norm extraction unit 212 If the input norm value is not -1, it is determined as the norm.

  The norm comparison unit 217 compares the unit norm read from the unit norm storage unit 236 with the input norm read from the input norm storage unit 237 in response to the token input via the token input unit 202.

  The selection output unit 251 determines that the unit norm read from the unit norm storage unit 236 and the unit read from the unit network coordinate storage unit 234 when the comparison result of the norm comparison unit 217 is smaller than the input norm. When the unit norm output from the data output unit 204 as a pair with the network coordinates and read out is equal to or larger than the input norm, the input norm read from the input norm storage unit 237 and the input network coordinate storage unit 233 are read. The input network coordinates are combined and output via the data output unit 204. The selection output unit 251 includes a network coordinate selection unit 219, a norm selection unit 220, and a data synthesis output unit 221.

  When the comparison result of the norm comparison unit 217 indicates that the read unit norm is smaller than the input norm, the norm selection unit 220 reads the unit norm from the unit norm storage unit 236, and uses the read unit norm as the data synthesis output unit 221. Output to. Further, the network coordinate selection unit 219 reads the input norm from the input norm storage unit 237 when the comparison result of the norm comparison unit 217 indicates that the read unit norm is greater than or equal to the input norm, and synthesizes the read input norm Output to the output unit 221.

  When the result of comparison by the norm comparison unit 217 indicates that the read unit norm is smaller than the input norm, the network coordinate selection unit 219 reads the unit network coordinates from the unit network coordinate storage unit 234 and uses the read unit network coordinates as data. The result is output to the composite output unit 221. The network coordinate selection unit 219 reads the input network coordinates from the input network coordinate storage unit 233 when the unit norm read by the norm comparison unit 217 is greater than or equal to the input norm, and reads the input network The coordinates are output to the data composition output unit 221.

  The data synthesis output unit 221 synthesizes a unit norm input from the norm selection unit 220 or a norm that is an input norm and a network coordinate that is a unit network coordinate or input network coordinate input from the network coordinate selection unit 219. Then, the norm synthesized as a set and the network coordinates are output via the data output unit 204.

  When the input value input to the data input unit 201 is a set of network coordinates and proximity determination identification information that is identification information indicating that the proximity determination is executed, the proximity determination unit 215 stores the unit network coordinate storage. It is determined whether the unit network coordinates read from the unit 234 are in the vicinity of the input network coordinates that are the input network coordinates.

  Further, the neighborhood determination unit 215 receives the unit network coordinates read from the unit network coordinate storage unit 234 in response to the input of the neighborhood determination identification information from the input norm determination unit 250 or the input norm determination unit 218. It is determined whether or not the input network coordinates are coordinates.

  The proximity determination unit 215 stores the distance when the input value input to the data input unit 201 is a set of network coordinates and proximity determination identification information that is identification information indicating that the proximity determination is executed. The reference distance is read from the unit 232, and it is determined whether or not the distance between the unit network coordinates read from the unit network coordinate storage unit 234 and the input network coordinates that are the input network coordinates is equal to or less than the read reference distance. By doing so, it is determined whether or not the unit network coordinates are in the vicinity of the input network coordinates.

  Further, the proximity determination unit 215 reads a reference distance associated with the number of learning steps that matches the number of steps read from the counter 213 from the distance storage unit 232, and the unit network coordinates are calculated based on the read reference distance. It is determined whether it is in the vicinity of the input network coordinates.

  The weight update unit 216 updates the value of the coupling weight vector stored in the weight storage unit 231 when the result determined by the proximity determination unit 215 is a neighborhood. Further, the weight update unit 216 calculates the difference between the input data vector read from the input data storage unit 230 and the combined weight vector read from the weight storage unit 231, and reads the calculated difference from the learning speed storage unit 235. A value obtained by multiplying the learning speed coefficient is added to the connection weight vector stored in the weight storage unit 231, thereby updating the value of the connection weight vector stored in the weight storage unit 231.

  Also, the weight update unit 216 reads from the learning speed storage unit 235 the learning speed coefficient associated with the number of learning steps that matches the number of steps read from the counter 213, and the weight storage unit 231 based on the read learning speed coefficient. Update the value of the connection weight vector stored in.

  The token output flag storage unit 238 stores a token output flag indicating whether or not the token output unit 203 has already output a token. The token output flag determination unit 223 reads the token output flag from the token output flag storage unit 238 in response to the token input to the token input unit 202, and when the read token output flag has not output a token. The input token is output to the norm comparison unit 217, and the input token is output via the token output unit 203.

  The token output unit 203 outputs a token in response to the token input to the token input unit 202. The token output unit 203 also includes a data output unit that includes the unit norm read from the unit norm storage unit 236 and the unit network coordinates read from the unit network coordinate storage unit 234 by the data synthesis output unit 221 or the selection output unit 251 as a set. Output via the data output unit 204 in response to the output via the data output unit 204 or as a combination of the input norm read from the input norm storage unit 237 and the input network coordinates read from the input network coordinate storage unit 233 Depending on, the token is output. Further, after outputting the token, the token output unit 203 updates the token output flag stored in the token output flag storage unit 238 to be output.

<Unit operation when applied to a self-organizing map>
Next, with reference to FIGS. 10 and 11, the operation of the units of the parallel arithmetic device when applied to the self-organizing map will be described. First, the operation when an input value is input to the data input unit of the unit will be described with reference to FIG.

  First, data that is an input value is input to the data input unit 201 (step S400). Next, the data input determination unit 222 determines whether or not the norm value of the input value is −2 that is a value indicating that the input value is an input data vector. It is determined whether or not the input value is an input data vector (step S401).

  Next, when the determination result in step S401 is that the norm value is −2 and an input data vector, the data extraction unit 210 stores the input data vector input in the input data storage unit 230 ( A count-up signal is output to the counter 213 together with step S402). Next, the counter 213 counts up the number of steps in response to the input of the count up signal from the data extraction unit 210 (step S403).

  Next, the norm calculation unit 214 calculates a norm between the input data vector input via the data extraction unit 210 and the combined weight vector read from the weight storage unit 231, and uses the calculated norm as a unit norm. The data is stored in the storage unit 236 (step S404), and the process ends.

  On the other hand, if the determination result in step S401 is that the norm value is not −2 and is not an input data vector, the network coordinate extraction unit 211 extracts network coordinates from the input values that have been input, and extracts the extracted network. The coordinates are stored in the input network coordinate storage unit 233 as input network coordinates (step S405), and the norm extraction unit 212 extracts the norm from the input value that has been input, and outputs the extracted norm to the input norm determination unit 218. To do.

  Next, the input norm determination unit 218 determines whether or not the norm value input from the norm extraction unit 212 is −1 (step S406), and the determination result is the norm input from the norm extraction unit 212. If the value of is not −1, it is determined as a norm, the input norm is stored as an input norm in the input norm storage unit 237 (step S407), and the process is terminated.

  On the other hand, if the determination result in step S406 is that the norm value input from the norm extraction unit 212 is -1, the input norm determination unit 218 determines that the input norm is the proximity determination identification information. The proximity determination identification information is output to the proximity determination unit 215. Next, the neighborhood determining unit 215 inputs the unit network coordinates read from the unit network coordinate storage unit 234 in response to the input of the neighborhood determining identification information from the input norm determining unit 218 being the input network coordinates. It is determined whether or not it is in the vicinity of the network coordinates (step S408).

  If the determination result in step S408 is near, the weight updating unit 216 updates the value of the coupling weight vector stored in the weight storage unit 231 (step S409), and the process ends. On the other hand, if the determination result in step S408 is not close, the process ends.

  Next, the operation when a token is input to the token input unit of the unit will be described with reference to FIG. First, a token is input to the token input unit 202 (step S500). Next, the token output flag determination unit 223 reads the token output flag from the token output flag storage unit 238 in response to the token input to the token input unit 202, and the read token output flag indicates that the token has not been output. It is determined whether or not there is a token (step S501). If the read token output flag indicates that the token has not been output, the input token is output to the norm comparison unit 217, and the input token is output as a token. The data is output via the unit 203.

  Next, the norm comparison unit 217 calculates the unit norm read from the unit norm storage unit 236 and the input norm read from the input norm storage unit 237 in response to the token input via the token input unit 202. Compare (step S502).

  When the comparison result of the norm comparison unit 217 indicates that the read unit norm is smaller than the input norm, the norm selection unit 220 reads the unit norm from the unit norm storage unit 236, and the read unit norm is the data synthesis output unit 221. (Step S503), the network coordinate selection unit 219 reads the unit network coordinates from the unit network coordinate storage unit 234, and outputs the read unit network coordinates to the data composition output unit 221 (step S504).

  Next, the data combination output unit 221 combines the unit norm input from the norm selection unit 220 and the unit network coordinates input from the network coordinate selection unit 219 as a set, and combines the unit norm and unit network combined as a set. The coordinates are output via the data output unit 204 (step S505).

  On the other hand, if the result of comparison by the norm comparison unit 217 in step S502 indicates that the read unit norm is greater than or equal to the input norm, the norm selection unit 220 reads the input norm from the input norm storage unit 237 and reads the read input norm. Is output to the data composition output unit 221 (step S508), and the network coordinate selection unit 219 reads the input network coordinates from the input network coordinate storage unit 233, and outputs the read input network coordinates to the data composition output unit 221 ( Step S509).

  Next, the data synthesis output unit 221 synthesizes the input norm input from the norm selection unit 220 and the input network coordinates input from the network coordinate selection unit 219 as a set, and combines the input norm and the input network combined as a set. The coordinates are output via the data output unit 204 (step S510).

  Next, in step S <b> 505, the token output unit 203 outputs data by combining the unit norm read from the unit norm storage unit 236 and the unit network coordinates read from the unit network coordinate storage unit 234 by the data composition output unit 221. The data output unit 204 sets the input norm read from the input norm storage unit 237 and the input network coordinates read from the input network coordinate storage unit 233 as a set in response to the output via the unit 204 or in step S510. In response to the output via, a token is output (step S506).

  Next, after outputting the token, the token output unit 203 updates the token output flag stored in the token output flag storage unit 238 to have been output (step S507), and ends the process.

<Operation of entire parallel computing device when adapted to self-organizing map>
Next, the operation of the entire parallel computing device when applied to the self-organizing map will be described. First, the control device inputs an input value composed of an input vector and a norm having a value of −2 to all units via the unit input bus 3 (procedure A300).

  Next, the norm calculation unit 214 of each unit that has received the input vector calculates the norm between the input vector that has been input and the connection weight vector that the unit has, and stores the calculated norm in the unit norm storage unit 236. (Procedure A301).

  Next, the control device inputs a token to the unit U1 (procedure A302). Next, the norm comparison unit 217 of the unit U1 receives the unit norm read from the unit norm storage unit 236 and the input norm storage unit 237 in response to the token input via the token input unit 202 of the unit U1. The read input norm is compared (procedure A303).

  Here, the input norm storage unit 237 of the unit U1 will be described assuming that a norm having a sufficiently large value is stored in advance as compared with the norm calculated by the unit U1 in step A301.

  Here, since the unit norm read from the unit norm storage unit 236 is smaller than the input norm read from the input norm storage unit 237, the selection output unit 251 of the unit U1 reads the unit norm read from the unit norm storage unit 236. And the unit network coordinates read from the unit network coordinate storage unit 234 are output as a set via the data output unit 204 (procedure A304).

  Next, the norm output from the data output unit 204 of the unit U1 and the network coordinates are input as a set to the data input unit 201 of each unit via the unit output bus 2, the amplifier 4, and the unit input bus 3. (Procedure A305).

  Next, since the norm value of the input value input to the data input unit 201 is not −1, the input norm unit 250 of each unit stores the input network coordinates as input network coordinates in the input network coordinate storage unit 233. The input norm is stored in the input norm storage unit 237 as an input norm (procedure A306). Here, the network coordinates stored in the input network coordinate storage unit 233 and the norm stored in the input norm storage unit 237 are the network coordinates and norm of the unit U1.

  Next, the token output unit 203 of the unit U1 includes the unit norm read from the unit norm storage unit 236 and the unit network coordinates read from the unit network coordinate storage unit 234 by the data synthesis output unit 221 or the selection output unit 251 of the unit U1. And a token is output to the unit U2 via the daisy chain control bus 1 (procedure A307).

  Next, similarly to the unit U1 in the procedure A303, the norm comparison unit 217 of the unit U2 receives the token from the unit U1 via the token input unit 202 of the unit U2, and then the unit norm storage unit 236. Is compared with the input norm read from the input norm storage unit 237 (procedure A308).

  Here, since the input norm stored in the input norm storage unit 237 is the norm of the unit U1, and the unit norm read from the unit norm storage unit 236 is the norm of the unit U2, the norm of the unit U2 The comparison unit 217 compares the norm of the unit U2 with the norm of the unit U1. Thereafter, based on the comparison result of the norm comparison unit 217 of the unit U2, the processing from the procedure A304 to the procedure A306 is executed similarly (procedure A309).

  Therefore, when the token is input to the unit U2, the norms of the unit U1 and the unit U2 are compared, the unit with the smaller norm is selected as the winner unit, and the winner unit norm and unit network coordinates are all Are stored in the input norm storage unit 237 and the input network coordinate storage unit 233 of the unit (step A310).

  Next, similarly to the procedure A307, the token is transmitted from the unit U2 to the unit U3, and then the unit U3 executes the procedure A303 to the procedure A307, thereby winning the unit U1 and the unit U2. The norm of the unit is compared with the norm of the unit U3, and the winner unit is selected from the units U1 to U3 (procedure A311).

  Thereafter, each time tokens are delivered to the units in the daisy chain order, the winner unit among the units to which the tokens are input is selected (procedure A312). Thereafter, when a token is input to the unit U10 which is the last unit in the daisy chain order, the data output unit 204 of the unit U10 outputs the norm and network coordinates of the winner unit in the unit U10 from the unit U1 ( Procedure A313).

  The control device receives the norm and network coordinates output from the data output unit 204 of the unit U10 as the norm and network coordinates of the winner unit (procedure A314). Thereby, the control apparatus can receive the norm of the winner unit and the network coordinates with respect to the input vector.

  Next, the control device sends the input value that is a combination of the network coordinates received from the data output unit 204 of the unit U10 and the norm with a value of −1 to all units via the unit input bus 3. Input (procedure A315).

  Next, each unit to which an input value that is a combination of the network coordinate and the norm with a value of −1 is input, the network coordinate extraction unit 211 of each unit extracts the network coordinate from the input value. The extracted network coordinates are stored in the input network coordinate storage unit 233 as input network coordinates (procedure A316), and the input norm determination unit 218 of each unit has a norm value of −1 input from the norm extraction unit 212. Therefore, it is determined that the input norm is the proximity determination identification information, and the proximity determination identification information is output to the proximity determination unit 215 (procedure A317).

  Next, the unit network coordinates read from the unit network coordinate storage unit 234 by the proximity determination unit 215 of each unit in response to the input of the proximity determination identification information from the input norm determination unit 250 or the input norm determination unit 218, It is determined whether or not the input network coordinate is read from the input network coordinate storage unit 233 (step A318).

  Next, the weight update unit 216 of each unit updates the value of the combined weight vector stored in the weight storage unit 231 of the unit when the result determined by the proximity determination unit 215 of the unit is a neighborhood. (Procedure A319). Thereby, the connection weight vector of the unit that is in the vicinity of the network coordinates of the winner unit is updated, and learning is executed.

  In the parallel computing device adapted to the self-organizing map described above, all the units are connected to the unit input bus 3, so that the control device only outputs the input vector to the unit input bus 3 once. Thus, input vectors can be input to all units. Further, since each unit can calculate the norm in parallel with respect to the input vector that has been input, the norm can be calculated in a short time for the entire unit.

  Further, since all the units are connected to the unit input bus 3, the control device only outputs the network coordinates of the winner unit once to the unit input bus 3, and inputs the network coordinates of the winner unit to all the units. be able to. In addition, since each unit determines and learns in parallel whether or not each unit is in the vicinity of the network coordinates of the input winner unit, the entire unit can be learned in a short time.

  Also, by passing tokens between units via the daisy chain control bus 1, winner units can be determined in order of daisy chain. That is, the winner unit can be determined by simple wiring of the daisy chain control bus 1.

<Parallel processing unit with redundancy>
Next, the configuration of the parallel arithmetic unit when redundancy is provided to the daisy chain control bus 1 will be described with reference to FIG.
Each unit has a first redundancy order that is a skipping order in a daisy chain order that is a predetermined order, and half of the plurality of units are connected to the first redundant daisy chain control bus 6. Daisy chained by. For example, the unit U 1, the unit U 3, the unit U 5, the unit U 7 and the unit U 9 are daisy chained by the first redundant daisy chain control bus 6.

  Further, in the daisy chain order that is a predetermined order, the remaining half of the plurality of units that are not daisy chained by the first redundant daisy chain control bus 6 in the order of one skip are The two redundant daisy chain control buses 7 are daisy chained. For example, the unit U 2, the unit U 4, the unit U 6, the unit U 8, and the unit U 10 are daisy chained by the second redundant daisy chain control bus 7.

  Next, the configuration of units daisy-chained by the first redundant daisy chain control bus 6 will be described with reference to FIG. That is, for example, the configuration of the unit U1, the unit U3, the unit U5, the unit U7, and the unit U9 will be described. In order to describe only the configuration that is changed with respect to redundancy, only the configuration that is changed in the configuration of the unit in FIG. 2 will be described.

  In FIG. 13, FIGS. 13A and 13C show the configuration of the unit shown in FIG. 2 before being changed. By making the parallel computing device have redundancy, the configuration of the unit in FIG. 13A is changed to the configuration in FIG. Further, the configuration of the unit in FIG. 13C is changed to the configuration of the unit in FIG.

  As shown in FIG. 13B, the unit includes a token output unit A13_1, a token output unit B13_2, a token output unit selection unit 17, and a token selection information storage unit 18. Here, the token output unit A13_1 corresponds to the token output unit 13 of FIG.

  The token output unit B13_2 (redundant token output unit) sends the token to the next unit in the redundant order via the first redundant daisy chain control bus 6 in response to the token input to the token input unit 12. Output. The token selection information storage unit 18 stores token output selection information indicating which one of the token output unit A13_1 and the token output unit B13_2 is selected.

  Based on the token output selection information read from the token selection information storage unit 18 in response to the token input to the token input unit 12, the token output unit selection unit 17 performs the token output unit A13_1 and the token output unit B13_2. Is selected, and a token is output via the selected token output unit A13_1 or token output unit B13_2.

  Moreover, as shown in FIG.13 (d), a unit has token input part A12_1 and token input part B12_2 (redundant token input part). Here, the token input unit A12_1 corresponds to the token input unit 12 of FIG.

The token input unit B12_2 inputs a token from the unit in the previous redundancy order that is daisy chained via the first redundant daisy chain control bus 6.
Further, the token output unit 13 outputs a token in response to the token input to the token input unit A12_1 or the token input unit B12_2.

  Similarly to the units daisy chained by the first redundant daisy chain control bus 6, the units daisy chained by the second redundant daisy chain control bus 7 also use the second redundant daisy chain control bus 7. The first redundant daisy chain control bus 6 includes a token input unit B12_2 (redundant token output unit), a token selection information storage unit 18, a token output unit selection unit 17, and a token input unit B12_2 (redundant token input unit). Yes.

  Here, the token selection information storage unit 18 has been described assuming that the token output selection information indicating which one of the token output unit A13_1 and the token output unit B13_2 is selected is stored. The token output selection information may be information indicating whether or not the next unit is defective in the daisy chain order, for example.

  When the token output selection information is information indicating whether or not the next unit is defective in the daisy chain order, the token output unit selection unit 17 is based on the token output selection information read from the token selection information storage unit 18. If there is no defect in the next unit in the daisy chain order, the token output unit A13_1 is selected and a token is output. Conversely, if there is a defect, the token output unit B13_2 is selected and a token is output. To do.

<Defect check method>
Next, a defect check method will be described. Here, a case where the parallel arithmetic device is a hierarchical neural network will be described. First, the parallel processing unit of the neural network executes the calculation as a hierarchical neural network having the same coupling coefficient and two layers of the input layer and the output layer without the intermediate layer, and the control device outputs the calculation result. Get the value. The control device determines the output value, and if normal, completes the process. When the control device determines the output value and detects that there is an abnormality, it writes defect information in the token selection information storage unit 18 of the preceding unit of the abnormal unit.

  Next, the determination method and its processing when the control device is normal and abnormal will be described in detail. First, when all the units obtain the same output from all units, the control device determines that the unit is normal. Next, the control device determines that the output from the unit stops in the middle or when the output value from the unit is different from the expected value as an abnormality. In this case, the control device writes the defect information in the token selection information storage unit 18 of the preceding unit that is the previous unit in the daisy chain order of the abnormal unit that is the unit detected as abnormal, performs the defect check again, The defect check is executed until the judgment result becomes normal. At this time, a mechanism for electrically disconnecting the circuit may be provided by providing a high-impedance switch in the input / output unit of the unit in case the electrical disconnection is required due to a defect.

  As described above, the parallel arithmetic device has redundancy by using the first redundant daisy chain control bus 6 and the second redundant daisy chain control bus 7, thereby enabling the daisy chain control bus 1 or any one of them. Even when a defect occurs in the unit, the parallel arithmetic device can operate without any problem.

  Here, a case has been described in which the redundant wirings are the first redundant daisy chain control bus 6 and the second redundant daisy chain control bus 7 which are skipped by one in the daisy chain order. It is possible to add two redundant daisy chain control buses.

<Second basic configuration>
Next, a second basic configuration of the parallel arithmetic device according to the embodiment of the present invention will be described with reference to the schematic block diagram of FIG. In the schematic block diagram of FIG. 14, the same components as those of the first basic configuration of FIG. Here, a case where the number of units of units AU1 to AU10 is ten will be described. Details of this unit will be described later.

  In the second basic configuration, the parallel arithmetic device has a plurality of units (AU1 to AU10) identified by unique unit numbers that are predetermined identification numbers, and any one of the plurality of units. Is a selection unit that is an identification number for identifying any one unit selected from the inputted input value and a plurality of units in response to the output value output by the unit being input via the unit output bus 2 And a distribution control unit 8 that outputs control data including a number to each of the plurality of units via the unit input bus 3. The control data includes an input value, a selected unit number, and a control code for selecting a unit calculation method or control method.

  Next, the configuration of each unit will be described with reference to FIG. Since units AU1 to AU10 have the same configuration, only the configuration of unit AU1 will be described here. The unit AU1 includes a data input unit 11A, a data output unit 14A, a unit calculation unit 15A, a unit output storage unit 16A, a control code extraction unit 17A, an input value extraction unit 18A, a unit number extraction unit 19A, It has a unit number coincidence determination unit 20A and a unique unit number storage unit 21A.

  The unique unit number storage unit 21A stores in advance a unique unit number that is an identification number for identifying each unit. The unit output storage unit 16A stores a calculation result that is a result of calculation by the unit calculation unit 15A. The data input unit 11A inputs control data from the distribution control unit 8 via the unit input bus 3. The input value extraction unit 18A extracts an input value from the control data input to the data input unit 11A. The selected unit number extraction unit 19A extracts the selected unit number from the control data input to the data input unit 11A. The control code extraction unit 17A extracts a control code from the control data input to the data input unit 11A.

  The unit number match determination unit 20A determines whether or not the output unit number included in the control data input to the data input unit 11A matches the unique unit number. The unit number match determination unit 20A determines whether the selected unit number extracted by the selected unit number extraction unit 19A matches the unique unit number read from the unique unit number storage unit 21A.

  The unit calculation unit 15A calculates by a calculation method predetermined for each unit based on the input value included in the control data input to the data input unit 11A. This unit calculation unit 15A calculates by a calculation method predetermined for each unit based on the input value extracted by the input value extraction unit 18A. The unit calculation unit 15A selects a calculation method from a predetermined calculation method based on the control code extracted by the control code extraction unit 17A, and performs the calculation using the selected calculation method. Further, the unit calculation unit 15A stores the calculation result that is the calculation result in the unit output storage unit 16A.

  When the result determined by the unit number coincidence determination unit 20A is coincident, the data output unit 14A uses the operation result obtained by the unit operation unit 15A as an output value as an output value, and sends the unit output bus 2 to the distribution control unit 8. Output via. In addition, when the result determined by the unit number coincidence determination unit 20A is coincident, the data output unit 14A reads the calculation result from the unit output storage unit 16A, and uses the read calculation result as an output value as the distribution control unit 8 Are output via the unit output bus 2.

  Next, the configuration of the distribution control unit 8 will be described with reference to FIG. The distribution control unit 8 includes a unit output value input unit 31, a control data generation unit 32, a control data output unit 33, and a processing procedure storage unit 34. The unit output value input unit 31 inputs an output value output from any one of the plurality of units via the unit output bus 2. The control data generation unit 32 generates control data by a predetermined processing procedure based on the output value input to the unit output value input unit 31. The control data output unit 33 outputs the control data generated by the control data generation unit 32 to each of the plurality of units via the unit input bus 3. The control data output unit 33 may include an amplifier that electrically amplifies the control data to be output.

  Further, the processing procedure storage unit 34 stores in advance processing procedures for executing calculations. The control data generation unit 32 sequentially reads out the processing procedure from the processing procedure storage unit 34, and if necessary, inserts the output value input to the unit output value input unit 31 into the read processing procedure. Control data is generated by a predetermined processing procedure.

  Here, the control data generation unit 32 of the distribution control unit 8 determines in advance in response to the output value being input to the unit output value input unit 31 or after the control data output unit 33 outputs the control data. Control data is generated after a predetermined time has elapsed. Therefore, the distribution control unit 8, more specifically, the control data output unit 33 of the distribution control unit 8 is controlled by the control data output unit 33 in response to the output value being input to the unit output value input unit 31. The control data is output after a predetermined time has elapsed since the data was output.

  Further, the distribution control unit 8 receives an input value which is data such as an initial value from the outside via the input terminal P2, and sets the input value to be input to each unit via the unit input bus 3. Execute parallel operations using each unit. In addition, the distribution control unit 8 receives the calculation result calculated by each unit from the unit via the unit output bus 2, and outputs the input calculation result to the outside via the output terminal P1. The term “external” as used herein refers to a control device or a computer outside the parallel arithmetic device.

  In the second basic configuration of the parallel arithmetic device described with reference to FIGS. 14 to 16, since the distribution control unit 8 controls each unit with control data, the first basic of the parallel arithmetic device shown in FIG. Compared to the configuration, the processing in each unit is simplified, and the daisy chain control bus 1 in which the units are connected in the daisy chain order is not required.

<Third Embodiment>
<Unit configuration when applied to hierarchical neural network>
Next, a configuration when the parallel arithmetic device according to the second basic configuration is applied to a hierarchical neural network will be described. Also in the hierarchical neural network, the configuration of the entire parallel arithmetic device is the same as the configuration of the parallel arithmetic device in FIG. Accordingly, only the configuration of the unit when applied to the hierarchical neural network will be described with reference to FIG.

  The unit includes a data input unit 101A, a data output unit 104A, a weight storage unit 130, a product-sum calculation primary value storage unit 131, a layer information storage unit 133, a function storage unit 132, and a unit output storage unit 135A. A preceding unit output storage unit 138B, a preceding unit number storage unit 139B, a selected unit number storage unit 141B, and a unique unit number storage unit 140B. The unit includes a product-sum calculation primary value calculation unit 111A, a unit output calculation unit 112A, a comparison unit 115B, a unit control unit 116B, a function number extraction unit 150B, a function number storage unit 151B, and an operand extraction unit. 152B and a general-purpose buffer storage unit 153B.

  Here, the data input unit 101A, the data output unit 104A, and the unit output storage unit 135A in FIG. 17 respectively correspond to the data input unit 11A, the data output unit 14A, and the unit output storage unit 16A in FIG. Also, the unique unit number storage unit 140B and the comparison unit 115A in FIG. 17 respectively correspond to the unique unit number storage unit 21A and the unit number match determination unit 20A in FIG. Descriptions of common functions in the configurations corresponding to those in FIGS. 17 and 15 are omitted.

  Further, the function number extraction unit 150B and the function number storage unit 151B in FIG. 17 correspond to the control code extraction unit 17A in FIG. Further, the operand extraction unit 152B and the general-purpose buffer storage unit 153B in FIG. 17 correspond to the input value extraction unit 18A or the selection unit number extraction unit 19A in FIG.

  Also, the weight storage unit 130, the layer information storage unit 133, the product-sum calculation primary value storage unit 131, the function storage unit 132, the product-sum calculation primary value calculation unit 111A, and the unit output calculation unit 112A illustrated in FIG. The preceding unit number storage unit 139B corresponds to the unit calculation unit 15A in FIG.

  In the configuration of FIG. 17, the same reference numerals or the same reference numerals are used for the same configuration as the configuration of the units of the parallel arithmetic unit when applied to the hierarchical neural network in the first basic configuration of FIG. 3. The code | symbol which attached | subjected code | symbol A is attached, and the description is abbreviate | omitted.

<Description of Each Configuration of Unit in Third Embodiment>
The function number storage unit 151B stores a control code (function number) included in the control data. In the general-purpose buffer storage unit 153B, information other than the control code A (function number) included in the control data, for example, a selection unit number B, an output value C, distribution layer information D, or distribution unit identification information E described later is stored. Remembered. Hereinafter, information other than the control code A (function number) included in the control data, for example, a selection unit number B, an output value C, distribution layer information D, or distribution unit identification information E described later will be described as general-purpose data. . Details of the control data will be described later with reference to FIG.

  The function number extraction unit 150B extracts the control code included in the control data input to the data input unit 101A, and stores the extracted control code in the function number storage unit 151B. The operand extraction unit 152B extracts general-purpose data included in the control data input to the data input unit 101A, and stores the extracted general-purpose data in the general-purpose buffer storage unit 153B.

  The unit control unit 116B executes processing based on the control data input to the data input unit 101A and the result determined by the comparison unit 115B, or selectively performs processing from among predetermined processing units. Let it run. The control data input to the data input unit 101A is a control code read from the function number storage unit 151B and general-purpose data read from the general-purpose buffer storage unit 153B.

  When the control code included in the control data input to the data input unit 101A indicates that the unit-sum calculation primary value is to be initialized, the unit control unit 116B stores the product-sum calculation primary value storage unit 131. The value of the stored product-sum calculation primary value information is initialized to zero.

  Further, the unit control unit 116B indicates that the control code included in the control data input to the data input unit 101A calculates a product-sum calculation primary value and adds and stores the result, and the control data outputs the output value and the output value. When the distribution unit identification information that is the unit identification information of the unit that is outputting is included, the distribution unit identification information is stored in the preceding unit output storage unit 138B, and the product-sum calculation primary value calculation unit 111A based on the output value The process is selectively executed using.

  In addition, when the control code included in the control data input to the data input unit 101A indicates that the unit output calculation is executed, the unit control unit 116B selectively performs processing using the unit output calculation unit 112A. Execute.

  The unit control unit 116B indicates that the control code included in the control data input to the data input unit 101A outputs an output value as an active unit, and the control data includes a selected unit number. In addition, when the result determined by the comparison unit 115B (unit number match determination unit) is a match, the process is selectively executed using the process of the data output unit 104A, and the calculation result is sent from the unit output storage unit 135A. The read calculation result is output as an output value to the distribution control unit 8 via the data output unit 104A and the unit output bus 2.

  When the layer information and the operation result are stored in association with each other in the unit output storage unit 135A, the unit control unit 116B is a unit in which the control code included in the control data input to the data input unit 101A is active. When the control data includes the selected unit number and the result of determination by the comparison unit 115B is the same, the processing of the data output unit 104A is selectively performed. The calculation result corresponding to the layer information read from the layer information storage unit 133 is read from the unit output storage unit 135A, and the data output unit 104A and the unit output bus 2 are output to the distribution control unit 8 as output values. And output via.

In addition, when the control data input to the data input unit 101A includes distribution layer information, the unit control unit 116B displays the distribution layer information included in the control data input to the data input unit 101A as a layer information storage unit. 133 to store.
Specifically, when the control data input to the data input unit 101A includes distribution layer information, the unit control unit 116B extracts and extracts distribution layer information from the control data input to the data input unit 101A. The distribution layer information is stored in the layer information storage unit 133. Further, the unit control unit 116B extracts the distribution layer information from the control data input to the data input unit 101A by reading the distribution layer information from the general-purpose buffer storage unit 153B.

In addition, the unit control unit 116B inputs the distribution unit identification information, which is the unit identification information of the unit outputting the output value, into the data input unit 101A when the control data input to the data input unit 101A includes the distribution unit identification information. The distribution unit identification information included in the control data is stored in the preceding unit output storage unit 138B.
Specifically, the unit control unit 116B, when the control data input to the data input unit 101A includes distribution unit identification information that is unit identification information of a unit that outputs an output value, the data input unit 101A. The distribution unit identification information is extracted from the control data input to, and the extracted distribution unit identification information is stored in the unit number storage unit 139B. In addition, the unit control unit 116B extracts the distribution unit identification information from the control data input to the data input unit 101A by reading the distribution unit identification information from the general-purpose buffer storage unit 153B.

  The product-sum calculation primary value calculation unit 111A reads the combined load corresponding to the layer information read from the layer information storage unit 133 from the weight storage unit 130 according to the selection of the unit control unit 116B, The input value input to the data input unit 101A is multiplied, and the multiplied value is added to and stored in the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit 131.

  Further, the product-sum calculation primary value calculation unit 111A combines the unit identification information read from the previous unit output storage unit 138B and the layer information read from the layer information storage unit 133 according to the selection of the unit control unit 116B. The load is read from the weight storage unit 130, the read combined load is multiplied by the input value input to the data input unit 101A, and the multiplied value is stored in the product-sum calculation primary value storage unit 131. It is added to the product-sum calculation primary value information and stored.

  The unit output calculation unit 112A reads the product-sum calculation primary value information from the product-sum calculation primary value storage unit 131 according to the selection of the unit control unit 116B, and outputs an output function corresponding to the layer information read from the layer information storage unit 133 Is calculated from the function storage unit 132, the output value is calculated by substituting the read product sum calculation primary value information into the read output function, and the calculated output value is stored in the unit output storage unit 135A. Further, when the unit output calculation unit 112A stores the calculated output value in the unit output storage unit 135A, the unit output calculation unit 112A associates the calculated output value with the layer information read from the layer information storage unit 133, and stores the calculated output value in the unit output storage unit. Store in unit 135A.

<Protocol of Control Data in Third Embodiment>
Next, a control data protocol as an example will be described with reference to FIG. The control data includes a control code A, a selection unit number B, and an output value C, and further includes distribution layer information D and distribution unit identification information E. The distribution layer information D is layer information of a unit that outputs an output value C, and the distribution unit identification information E is unit identification information of a unit that outputs an output value C. The protocol of this control code is determined in advance as follows according to the value of the control code.

<Selection process 0>
When the value of the control code A included in the control data is “0” and is a control code indicating “initialize the product-sum calculation primary value”, the control data contains other information (general data ) Is not included. All units that have received the control data whose control code A value is “0” initialize the value of the product-sum calculation primary value stored in the product-sum calculation primary value calculation unit 111A to 0.

<Selection process 1>
In addition, when the value of the control code A included in the control data is “1” and the control code indicates that “the product-sum calculation primary value is calculated and added and stored”, the control data is an output value. C, distribution layer information D, and distribution unit identification information E. All the units that have received the control data whose control code A value is 1 have the layer information l as the distribution layer information D, the unit identification information i as the distribution unit identification information E, and the input value as the output value C. As a result, a product-sum calculation primary value is calculated.

<Selection process 2>
Further, when the value of the control code A included in the control data is “2” and the control code indicates that “unit output calculation is executed”, the control data includes the distribution layer information D. All the units that have received the control data whose control code A value is “2” execute the unit output calculation using the layer information l as the distribution layer information D, and the layer information l using the executed calculation result as the output value. And stored in the unit output storage unit 135A.

<Selection process 3>
When the value of the control code A included in the control data is “3” and the control code indicates that “output value is output as an active unit”, the control data includes the selected unit number B and Distribution layer information D. Whether all units that have received control data whose control code A value is “3” indicates whether or not the selected unit number B included in the received control data matches the unique unit number I of its own unit If the determined results match, the layer information l is set as the value of the distribution layer information D, and the output value read from the unit output storage unit 135A corresponding to the layer information l is output.

  Based on the control data protocol of FIG. 18, the distribution control unit 9 includes a control code indicating that the product-sum calculation primary value is initialized, an output value, and distribution unit identification information. A control code indicating that it is calculated and stored, a control code indicating that a unit output operation is executed, a control code indicating that an output value is output as an active unit, or control data including the distribution layer information, Output in a predetermined order according to the number of layers of the hierarchical neural network and the number of nodes in the layer.

<Operation of Unit in Third Embodiment>
Next, the operation of the unit, particularly the operation of the unit control unit 116 of the unit when the control data protocol described in FIG. 18 is received will be described with reference to FIG.
First, the unit starts processing in response to input of control data to the data input unit 101A (step S1901). Next, the function number extraction unit 150B extracts a control code from the control data input to the data input unit 101A, and stores the extracted control code in the function number storage unit 151B.

  Further, the operand extraction unit 152B receives information other than the control code that is the operand from the control data input to the data input unit 101A, for example, the selection unit number B, the output value C, the distribution layer information D, and the distribution unit. The identification information E is extracted, and the extracted operand is stored in the general-purpose buffer storage unit 153B.

  Next, based on the value of the control code read from the function number storage unit 151B, the unit control unit 116B extracts the control code from the function number extraction unit 150B, and will be described below from step S1902 to step S1908. Thus, the processing in the unit is selectively executed.

  First, the unit controller 116B determines whether or not the value of the read control code A is “0” (step S1902). If the determination result in step S1902 is that the value of the read control code A is “0”, the unit control unit 116B executes the selection process 0 described above (step S1903) and ends the process.

  On the other hand, if the determination result of step S1902 is that the value of the read control code A is not “0”, the unit control unit 116B determines whether the value of the read control code A is “1”. (Step S1904). If the determination result of step S1904 is that the value of the read control code A is “1”, the unit control unit 116B executes the selection process 1 described above (step S1905) and ends the process.

  On the other hand, if the determination result of step S1904 is that the value of the read control code A is not “1”, the unit control unit 116B determines whether the value of the read control code A is “2”. (Step S1905). If the determination result of step S1905 is that the value of the read control code A is “2”, the unit control unit 116B executes the selection process 2 described above (step S1906) and ends the process.

  On the other hand, if the determination result of step S1905 indicates that the value of the read control code A is not “2”, that is, if the value of the read control code A is “3”, the comparison unit 115B selects It is determined whether or not the selected unit number read from the unit number storage unit 141B matches the unique unit number read from the unique unit number storage unit 140B (step S1907). If the result of determination in step S1907 indicates that the selected unit number and the unique unit number match, the unit control unit 116B executes the selection process 3 described above (step S1908) and ends the process. On the other hand, if the determination result in step S1907 indicates that the selected unit number does not match the unique unit number, the process ends.

<Operation of Distribution Control Unit 8 in Third Embodiment>
Next, the operation of the distribution control unit 8 when the parallel arithmetic device having the second basic configuration is applied to a hierarchical neural network will be described with reference to FIGS. Since the operation of each unit according to the value of the control code included in the control data output by the distribution control unit 8 has already been described with reference to FIGS. 17 to 19, the operation of the distribution control unit 8 is here described. Only that will be described. Here, the case where the parallel arithmetic device having the second basic configuration is applied to the hierarchical neural network of FIG. 6 will be described.

  First, an outline of the operation of the distribution control unit 8 when applied to a hierarchical neural network will be described with reference to FIG. First, initial setting (initialization) of each unit is performed (step S2001). Next, in response to an input instruction which is input data (pattern input) for the node of the input layer L1 from the outside via the input terminal P2, each input as each node of the intermediate layer (hidden layer) L2 is received. When the distribution control unit 8 distributes the units as the input layer L1, the product-sum calculation of the intermediate layer L2 is executed in parallel using each unit (step S2002). In step S2002, the distribution control unit 8 outputs an output value to each unit serving as a node of the intermediate layer L2 as a node of the input layer L1.

  Next, the intermediate layer product sum calculation (step S2004) is repeated for all intermediate layers (layer information counter variable l = 2,..., L + 1) of the hierarchical neural network (layer loop from step L2002 to step L2005). processing). Here, L is the value of the total number of intermediate layers (total number of intermediate layers). In the hierarchical neural network having three layers in FIG. 6, the value of the total number L of intermediate layers is 1. The intermediate layer product sum calculation in step S2004 will be described in detail later with reference to FIG.

  Through the layer loop processing from step L2002 to step L2005, the processing as the intermediate layer node of the hierarchical neural network is executed in order from the input layer L1 to the output layer L3 in the layer of the hierarchical neural network. In each unit, the value of the operation result as the output layer L3 is stored. In the loop from step L2002 to step L2005, any one unit functions as an active unit serving as an output layer, and all units function as nodes in a layer receiving input.

  Next, the output values are output to the units as the nodes of the output layer L3 in the order of the units identified by the unit identification information (step S2006), whereby all the nodes that belong to the output layer L3 of the hierarchical neural network are output. The calculation results from the unit are received in order. In addition, the calculation results received in this order are sequentially output to the outside through the output terminal P1 as the calculation results of the hierarchical neural network calculation device for the input pattern input.

  Through the processing from step S2001 to step S2006, the parallel arithmetic device performs the calculation of the hierarchical neural network in parallel using a plurality of units according to the input which is input data (pattern input) input from the outside. It can be executed and the result of the executed operation can be output to the outside.

  Next, each process of FIG. 20A will be described in detail with reference to FIGS. 20B, 20C, 21A, and 21B. Here, the processing of FIG. 20B, FIG. 20C, FIG. 21A, and FIG. 21B will be described as functions or subroutines for each processing of FIG. .

  First, the process in step S2001 in FIG. 20A will be described in detail with reference to FIG. First, control data with the value of the control code A set to “0” is output (step S20011), and then the process returns.

Next, the process in step S2002 in FIG. 20A will be described in detail with reference to FIG. First, the process of the next step S20022 and step S20023 is repeated while sequentially increasing the value of the counter variable i from 0 to 1, 2,... N ⁽¹⁾ (input loop of steps L20021 and L20024). . N ⁽¹⁾ is the value of the number of nodes in the layer whose layer identification information value is 1, that is, the value of the number of nodes in the input layer L1. For example, in the hierarchical neural network of FIG. 6, the value of the number of nodes in the input layer L1 is 3.

In step S20022, pattern input x _i ^(l) is input from the outside. Next, in step S20023, the value of the control code A is set to “1”, the value of the output value C is set to the pattern input x _i ⁽¹⁾ input from the outside in step S20022, and the value of the distribution layer information D is counted as a counter. Control data in which the variable i is set to 1 and the value of the distribution unit identification information E is 1 is output to each unit.
Next, control data with the value of the control code A set to “2” and the value of the distribution layer information D set to 1 is output to each unit (step S20025), and the process returns.

Next, the processing in step S2004 in FIG. 20A will be described in detail with reference to FIG. First, while sequentially increasing the value of the counter variable i from 0 to 1, 2,... N ^(l) , the processing from the next step S20042 to step S20044 is repeated (input loop of step L20041 and step L20045). N ^(l) is the value of the number of nodes in the layer whose layer identification information value is the layer information counter variable l.

In this step S20042, control data with the value of the control code A set to “3”, the selected unit number B as the counter variable i, and the distribution layer information D as the layer information counter variable l is output to each unit. In step S20043, the output value x _i ^(l) output from the unit in which the counter variable i, which is the selected unit number B output in step S20042, matches the unique unit number is input. In step S20044, the value of the control code A is “2”, the output value C is the output value x _i ^(l) input from the unit in step S20043, the distribution layer information D is the counter variable i, and the distribution is performed. Control data with the unit identification information E as the layer information counter variable l is output to each unit.
Next, control data in which the value of the control code A is “2” and the distribution layer information D is the layer information counter variable l is output to each unit (step S20046), and the process returns.

Next, the processing in step S2006 in FIG. 20A will be described in detail with reference to FIG. First, while sequentially increasing the value of the counter variable i from 0 to 1, 2,... N ^{(L + 2)} , the processing from the next step S20062 to step S20064 is repeated (the output loop of step L20061 and step L20065). Return processing. N ^{(L + 2)} is the value of the number of nodes in the layer whose layer identification information value is (intermediate layer total number L + 2), that is, the number of nodes in the output layer L3. For example, in the hierarchical neural network of FIG. 6, the value of the number of nodes of the output layer L3 is 3.

In this step S20062, control data in which the value of the control code A is “3”, the selected unit number B is the counter variable i, and the distribution layer information D is (intermediate layer total number L + 2) is output to each unit. In step S20063, the output value x _i ^{(L + 2)} output from the unit in which the counter variable i, which is the selected unit number B output in step S20062, matches the unique unit number is input. In step S20064, the output value x _i ^{(L + 2)} input from the unit in step S20063 is output to the outside via the output terminal P1.

  As a result of the above processing, the parallel computing device when applied to the hierarchical neural network according to the third embodiment is similar to the hierarchical computing device adapted to the hierarchical neural network according to the first embodiment. A neural network can be executed, and the same effects as the hierarchical neural network according to the first embodiment can be obtained.

<Fourth Embodiment>
<Unit structure when adapted to the self-organizing map>
Next, a configuration when the parallel arithmetic device according to the second basic configuration is applied to the self-organizing map will be described. Also in the self-organizing map, the configuration of the entire parallel arithmetic device is the same as the configuration of the parallel arithmetic device in FIG. Accordingly, only the configuration of the unit when applied to the self-organizing map will be described with reference to FIG.

  Here, in the parallel arithmetic unit of the self-organizing map in the first basic configuration, the unit having the minimum norm is determined by each unit calculating the norm, and then the norms are compared in order between the units. Was. On the other hand, in the parallel computing device of the self-organizing map in the second basic configuration, after each unit calculates the norm, the norm calculated by each unit is sequentially changed based on the control of the distribution control unit 8. The data is output to the distribution control unit 8 via the unit output bus 2. Next, the distribution control unit 8 determines a unit having the minimum norm by selecting the minimum norm from the input norms.

  Further, in the self-organizing map parallel computation device according to the first basic configuration, each unit associates the learning speed and the neighborhood distance with the number of times of learning in advance in an internal storage unit (learning rate storage unit and neighborhood distance storage unit ) On the other hand, in the parallel computing device of the self-organizing map in the second basic configuration, the distribution control unit 8 associates the learning speed with the learning speed storage unit and the neighborhood distance storage unit included therein in association with the number of learnings. The neighborhood distance is stored. Each unit stores the learning speed and the neighborhood distance of one value in an internal storage unit (learning rate storage unit and neighborhood distance storage unit). In addition, the distribution control unit 8 distributes the learning speed storage unit and the neighborhood distance storage unit included in the control data to each unit according to the number of learning, and each unit has an internal storage unit (learning speed storage unit and The learning speed storage unit and the proximity distance storage unit of the proximity distance storage unit) are updated to the learning speed storage unit and the proximity distance storage unit included in the control data received from the distribution control unit 8.

  As described above, the distribution control unit 8 distributes the learning speed storage unit and the neighborhood distance storage unit to each unit by including the control data in accordance with the number of learnings, so that all units can learn the number of times of learning. There is an effect that it is not necessary to store the learning speed and the neighborhood distance in advance in the internal storage unit. Also, for example, when changing the method of change between the learning speed and the neighborhood distance related to the number of learning, such as increasing or decreasing the degree of change between the learning speed and the neighborhood distance according to the number of learning. Only the information in the storage unit inside the distribution control unit 8 storing the learning speed storage unit and the neighborhood distance storage unit in association with the number of learnings need be changed, and the change can be simplified.

  Further, the unit when adapted to the self-organizing map is identified in advance by coordinates identified by network coordinates, as in the second embodiment. Therefore, in the fourth embodiment, the unique unit number will be described as network coordinates. Since each unit has the same configuration as in the second or third embodiment, only the configuration of one unit will be described here with reference to FIG.

  The unit includes a data input unit 201A, a data output unit 204A, a distance storage unit 232A, a weight storage unit 231A, a norm calculation unit 214A, a proximity determination unit 215A, a weight update unit 216A, and a unit network coordinate storage unit. 234A, input network coordinate storage unit 233A, learning speed storage unit 235A, unit norm storage unit 236A, input data storage unit 230A, real vector dimension number storage unit 242B, comparison unit 215B, and unit control unit 216B A function number extraction unit 250B, a function number storage unit 251B, an operand extraction unit 252B, and a general-purpose buffer storage unit 253B.

  Here, the data input unit 201A, the data output unit 204A, the unit norm storage unit 236A, and the weight storage unit 231A in FIG. 22 correspond to the data input unit 11A, the data output unit 14A, and the unit output storage unit 16A in FIG. To do. Further, the unit network coordinate storage unit 234A and the comparison unit 215B in FIG. 22 respectively correspond to the unique unit number storage unit 21A and the unit number match determination unit 20A in FIG. Descriptions of common functions in the configurations corresponding to those in FIGS. 22 and 15 are omitted.

  Further, the function number extraction unit 250B and the function number storage unit 251B in FIG. 22 correspond to the control code extraction unit 17A in FIG. The operand extraction unit 252B and the general-purpose buffer storage unit 253B in FIG. 22 correspond to the input value extraction unit 18A or the selection unit number extraction unit 19A in FIG.

  In the configuration of FIG. 22, the same reference numerals are given to the same configuration as the configuration of the units of the parallel arithmetic device when applied to the self-organizing map in the first basic configuration of FIGS. 7 and 8, or The same reference numerals are denoted by the reference numerals A, and the description thereof is omitted.

<Description of Each Configuration of Unit in Fourth Embodiment>
The function number storage unit 251B stores a control code (function number) included in the control data. In the general-purpose buffer storage unit 253B, information other than the control code A (function number) included in the control data, for example, an input network coordinate B, an input learning speed C, an input proximity distance D, an input weight W, and an input dimension number described later. N or the input value X is stored.

  Hereinafter, information other than the control code A (function number) included in the control data, for example, input network coordinates B, input learning speed C, input proximity distance D, input weight W, input dimension number N, or input described later The value X will be described as general-purpose data. Details of the control data will be described later with reference to FIG.

  The function number extraction unit 250B extracts the control code included in the control data input to the data input unit 201A, and stores the extracted control code in the function number storage unit 251B. The operand extraction unit 252B extracts general-purpose data included in the control data input to the data input unit 201A, and stores the extracted general-purpose data in the general-purpose buffer storage unit 253B.

  The unit control unit 216B executes processing based on the control data input to the data input unit 201A and the result determined by the comparison unit 215B, or selectively performs processing from among predetermined processing units. Let it run. The control data input to the data input unit 201A is a control code read from the function number storage unit 251B and general-purpose data read from the general-purpose buffer storage unit 253B.

  Also, the distance storage unit 232A, the norm calculation unit 214A, the proximity determination unit 215A, the weight update unit 216A, the input network coordinate storage unit 233A, the learning speed storage unit 235A, and the input data storage unit 230A illustrated in FIG. The real vector dimension number storage unit 242B and the unit control unit 216B correspond to the unit calculation unit 15A in FIG. The weight storage unit 231A in FIG. 22 corresponds to the unit calculation unit 15A and the unit output storage unit 16A in FIG.

  Details and other configurations of the unit control unit 216B will be described with reference to a control data protocol as an example with reference to FIG.

<Protocol of Control Data in Fourth Embodiment>
Next, an exemplary control data protocol will be described with reference to FIG. This control data includes control code A, input network coordinates B, input learning speed C, input proximity distance D, input coupling weight vector W (W1, W2,...), Input dimension number N, Input vector X (X1, X2,...). The protocol of this control code is determined in advance as follows according to the value of the control code.

<Selection process 4>
When the value of the control code A included in the control data is “4”, indicating that “weight is set”, the control data includes the input network coordinate B and the input coupling weight vector W (W1, W2, ...). For all units that have received the control data indicating that the value of the control code A is “4” and “set the weight”, the unit control unit 216B of the unit is stored in the input network coordinate storage unit 233A. The input network coordinate value is updated to the input network coordinate B included in the received control data, and the updated input network coordinate and the unit network coordinate read from the unit network coordinate storage unit 234A are compared by the comparison unit 215B. It is determined whether or not they match. Next, when the determination results of the comparison unit 215B match, the unit control unit 216B uses the combination weight vector stored in the weight storage unit 231A as the input combination weight vector W (W1) included in the received control data. , W2,...

<Selection process 5>
When the value of the control code A included in the control data is “5”, indicating that “the learning speed and the neighborhood distance are set”, the control data includes the input learning speed C and the input neighborhood distance D. And. The unit control unit 216B of the unit receives the control data indicating that the value of the control code A is “5” and “sets the learning speed and the neighborhood distance”. The learning speed storage unit 235A Is updated to the input learning speed C included in the received control data, and the reference distance stored in the distance storage unit 232A is updated to the input neighborhood distance D included in the received control data. .

<Selection process 6>
When the value of the control code A included in the control data is “6”, which indicates that “the number of dimensions of the real vector is set”, the control data includes the number of input dimensions N. The unit control unit 216B of the unit receives the control data indicating that the value of the control code A is “6” and “sets the dimension number of the real vector”. The number of dimensions stored in 242B is updated to the number N of input dimensions included in the received control data.

<Selection process 7>
When the value of the control code A included in the control data is “7”, indicating that “norm is calculated”, this control data includes the input vector X (X1, X2,...). Yes. For all units that have received the control data indicating that the value of the control code A is “7” and “calculate the norm”, the unit control unit 216B of the unit is stored in the input data storage unit 230A. The input data vector is updated to the input vector X (X1, X2,...). Next, the norm calculation unit 214A calculates a norm between the updated input vector X (X1, X2,...) And the combined weight vector read from the weight storage unit 231A, and the calculated norm is a unit norm. Is stored in the unit norm storage unit 236A.

<Selection process 8>
When the value of the control code A included in the control data is “8”, which indicates “output the norm”, the control data includes the input network coordinate B. For all units that have received the control data indicating that the value of the control code A is “8” and “output the norm”, the unit control unit 216B of the unit is stored in the input network coordinate storage unit 233A. The input network coordinate value is updated to the input network coordinate B included in the received control data, and the updated input network coordinate and the unit network coordinate read from the unit network coordinate storage unit 234A are compared by the comparison unit 215B. It is determined whether or not they match. Next, when the determination results of the comparison unit 215B match, the unit control unit 216B reads the unit norm read from the unit norm storage unit 236A via the data output unit 204A and the unit output bus 2. 8 is output.

<Selection process 9>
When the value of the control code A included in the control data is “9”, which indicates “learn”, the control data includes the input network coordinate B. All units that have received the control data indicating that the value of the control code A is “9” and “learn” are input network coordinates stored in the input network coordinate storage unit 233A by the unit control unit 216B. Is updated to the input network coordinates B included in the control data. Next, the unit proximity determination unit 215A calculates the distance between the unit network coordinates read from the unit network coordinate storage unit 234A and the input network coordinates read from the input network coordinate storage unit 233A, and the calculated distance is By determining whether or not the distance is less than or equal to the distance read from the distance storage unit 232A, the unit network coordinates read from the unit network coordinate storage unit 234A are close to the input network coordinates read from the input network coordinate storage unit 233A. It is determined whether or not there is. Next, when the result determined by the proximity determination unit 215A is a neighborhood, the difference between the input data vector read from the input data storage unit 230A by the weight update unit 216A and the combined weight vector read from the weight storage unit 231A. And the value obtained by multiplying the calculated difference by the learning speed coefficient read from the learning speed storage unit 235A is added to the combined weight vector stored in the weight storage unit 231A, and stored in the weight storage unit 231A. Learning is performed by updating the existing connection weight vector.

<Selection process 10>
When the value of the control code A included in the control data is “10”, indicating that “learning result is output”, the control data includes the input network coordinate B. For all units that have received the control data indicating that the value of the control code A is “10” and “output the learning result”, the unit control unit 216B of the unit is stored in the input network coordinate storage unit 233A. The input network coordinate value updated to the input network coordinate B included in the received control data, and the comparison unit 215B includes the updated input network coordinate and the unit network coordinate read from the unit network coordinate storage unit 234A. It is determined whether or not. Next, when the determination results of the comparison unit 215B match, the unit control unit 216B uses the data output unit 204A and the unit output bus 2 to transmit the combined weight vector read from the weight storage unit 231A to the distribution control unit. 8 is output.

<Operation of Unit in Fourth Embodiment>
Next, with reference to FIG. 24, the operation of the unit when the control data protocol described in FIG. 23 is received, in particular, the operation of the unit control unit 216B of the unit will be described. For the sake of simplicity, the following description will be made using the value of the control code A. The meaning of the value of the control code A is the same as described above with reference to FIG.

  First, the unit starts processing in response to input of control data to the data input unit 201A (step S2401). Next, the function number extraction unit 250B extracts the control code from the control data input to the data input unit 201A, and stores the extracted control code in the function number storage unit 251B.

  Further, the operand extraction unit 252B causes the general-purpose buffer storage unit 253B to store information other than the control code that is the operand from the control data input to the data input unit 201A.

  Next, based on the value of the control code read from the function number storage unit 251B in response to the function number extraction unit 250B extracting the control code, the unit control unit 216B performs steps S2402 to S2412 described below. Thus, the process in the unit is selectively executed (step S2402).

  If the read control code value is “4” in step S2402, the comparison unit 215B reads the unit network coordinates read from the unit network coordinate storage unit 234A and the input network coordinates read from the input network coordinate storage unit 233A. Is matched (step S2403). If the result of determination in step S2403 is that the selected unit number matches the unique unit number, the unit control unit 216B executes the selection process 4 described above (step S2404) and ends the process. On the other hand, if the determination result of step 2403 indicates that the selected unit number does not match the unique unit number, the process is terminated.

  On the other hand, if the value of the read control code is “5” in step S2402, the unit control unit 216B executes the selection process 5 described above (step S2405) and ends the process. On the other hand, if the value of the read control code is “6” in step S2402, the unit control unit 216B executes the selection process 6 described above (step S2406) and ends the process. On the other hand, if the value of the read control code is “7” in step S2403, the unit controller 216B executes the selection process 7 described above (step S2407) and ends the process.

  On the other hand, if the read control code value is “8” in step S2402, the comparison unit 215B inputs the unit network coordinates read from the unit network coordinate storage unit 234A and the input network coordinate storage unit 233A. It is determined whether or not the network coordinates match (step S2408). If the determination result of step S2408 indicates that the selected unit number and the unique unit number match, the unit control unit 216B executes the selection process 8 described above (step S2409), and ends the process. On the other hand, if the determination result of step 2408 does not match the selected unit number and the unique unit number, the process is terminated.

  On the other hand, if the value of the read control code is “9” in step S2402, the unit control unit 216B executes the selection process 9 described above (step S2410) and ends the process.

  On the other hand, if the read control code value is “10” in step S2402, the comparison unit 215B inputs the unit network coordinates read from the unit network coordinate storage unit 234A and the input network coordinate storage unit 233A. It is determined whether or not the network coordinates match (step S2411). If the result of determination in step S2411 is that the selected unit number matches the unique unit number, the unit controller 216B executes the selection process 10 described above (step S2412), and ends the process. On the other hand, if the result of determination in step 2411 does not match the selected unit number and the unique unit number, the process ends.

  As described above with reference to FIG. 24, each unit selectively executes predetermined processing based on the value of the control code included in the input data, by the unit control unit 216B of the unit. To do.

<Operation of Distribution Control Unit 8 in Fourth Embodiment>
Next, the operation of the parallel computing device when the parallel computing device according to the second basic configuration is applied to the self-organizing map will be described with reference to FIGS. Note that the parallel computing device includes the distribution control unit 8 and a plurality of units, but the operation of each unit according to the value of the control code included in the control data output by the distribution control unit 8 is shown in FIGS. Since it has already been described by using, only the operation of the distribution control unit 8 will be described here.

  First, the operation of the distribution control unit 8 of the parallel computing device when applied to the self-organizing map will be described with reference to FIG. First, initial setting of each unit is performed (step S2501). This initial setting will be described in detail later with reference to FIG.

  Next, competitive learning (step S2503) is repeated for the learning frequency variable t = 0, 1, 2,... T (learning loop processing from step L2502 to step L2504). Here, T is the number of times of competitive learning. The competitive learning in step S2502 will be described in detail later with reference to FIG.

  Next, the learned result is output (step S2505). This output will be described in detail later with reference to FIG.

  Next, with reference to FIG. 25B and FIGS. 26A to 26C, each process of FIG. 25A will be described in detail. Here, the processes in FIG. 25B and FIGS. 26A to 26C will be described as functions or subroutines for the processes in FIG. 20A.

  First, the initial setting process in step S2501 in FIG. 25A will be described with reference to FIG. First, the pattern input n is input from the outside (step S250101). Next, control data in which the value of the control code A is “6” and the value of the output value D is input to the pattern input n is output to each unit (step S250102).

  Next, in the order of the units indicated by unit loop variable j = 1, 2,... N, the processing of the next dimension loop from step L250104 to step L250106 and S250107 is repeated (unit loop from step L250103 to step L250108). . Here, N is a value of the total number of units.

  In the dimension loop from step L250104 to step L250106, the processing of the next step S250105 is repeated in the order of the dimension number of the input vector indicated by the input vector dimension number i = 1, 2,. In step S250105, an input pattern wji is input from the outside. Here, n is the number of dimensions of the input vector.

  In step S250107, the value of the control code A is “4”, the value of the input network coordinate B is the unit loop variable j, and the output value C (C1, C2,... Cn) is the input pattern wji (wj1). , Wj2,... Wjn) are output to each unit.

  By the processing from step L250103 to step L250108, the connection weight vector of the weight storage unit 231A in each unit is initialized for each unit.

Next, the processing in step S250110 is repeated for the learning frequency variable t = 0, 1, 2,... T (input loop from step L250109 to step L250111). In this step S250110, the learning speed η (t) and the neighborhood distance d (t) are input from the outside, and the learning speed η (t) input in association with the learning frequency variable t is stored in the internal learning speed storage unit. At the same time, the proximity distance d (t) input in association with the learning frequency variable t is stored in the internal proximity distance storage unit.
Next, the value of the norm minimum value dmin, which is a dummy variable used for minimum norm determination described later with reference to FIG. 26B, is set to −1 (step S250112), and the process returns.

  Next, the competitive learning process in step S2503 in FIG. 25A will be described with reference to FIG. First, the value of the control code A is “5”, the input learning speed C is the learning speed η (t) read from the internal learning speed storage unit, and the input neighborhood distance D is the neighborhood distance d read from the internal neighborhood distance storage unit. The control data set as (t) is output to each unit (step S250301).

  Next, the process of the next step S250303 is repeated for the dimension loop variable i = 1, 2,... N (the dimension loop from step L250302 to step L250304). In step S250303, a pattern input xi is input from the outside.

  Next, the value of the control code A is “7”, and the input vector C (C1, C2,... Cn) is input from the outside in steps L250302 to L250304, pattern input x (x1, x2,... Xn). The control data is output to each unit (step S250305).

  Next, for the unit loop number variable j = 1, 2,... N, the processing of the next step S250307, step S250308, and step S250309 is repeated (unit loop from step L250306 to step L250310). In step S250307, control data having the control code A value of “8” and the input network coordinate B value of the unit loop variable j is output to each unit. In step S250308, the unit network coordinates are the input network coordinates B for the control data in which the value of the control code A is “8” and the value of the input network coordinates B is the unit loop number variable j in step S250307. The norm output from the unit that matches a certain unit loop number variable j is input as norm dn.

  In step S250309, the minimum norm determination process is executed based on the norm dn input in step S250308. The minimum norm determination process in step S250309 will be described later in detail with reference to FIG.

  The norm calculated in each unit is input in order from all units by the processing in steps L250306 to L250310, and the norms input in the processing in step S250309 are compared in order, so that the norms from all units are compared. From the output, the minimum norm and the unit network coordinates that are identification information for identifying the unit that has output the minimum norm are selected as the norm minimum unit network coordinates jmin.

  Next, control data in which the value of the control code A is “9” and the value of the input network coordinate B is the norm minimum unit network coordinate jmin is output to each unit. Thereby, each unit learns based on the input norm minimum unit network coordinate jmin.

Next, the minimum norm determination process in step S250309 in FIG. 26A will be described with reference to FIG.
First, it is determined whether the norm dn input in step S250308 is smaller than the norm minimum value dmin or whether the value of the norm minimum value dmin is −1 (step S2503091). If the input norm dn is smaller than the norm minimum value dmin, or if the norm minimum value dmin is −1, the norm minimum value dmin is set as the input norm dn. The norm minimum unit network coordinate jmin is set as the unit loop variable j (step S2503092), and the process returns.

  On the other hand, if the result of the determination in step S2503091 is that the input norm dn is not smaller than the norm minimum value dmin, and the norm minimum value dmin is not −1, the process is returned.

  Next, the output process in step S2505 in FIG. 25A will be described with reference to FIG. First, for the unit loop variable j = 0, 1, 2,... N, the processing in the next step S250502 and the dimension loop processing from step L250503 to step L250506 are repeated (unit loop from step L250501 to step L250507).

  In step S250502, control data with the value of the control code A being “10” and the value of the input network coordinate B being the unit loop variable j is output to each unit.

  In the dimension loop process from step L250503 to step L250506, the process of the next step S250504 and step S250505 is repeated for the dimension loop variable i = 1, 2,... N. In step S250504, a connection weight vector that is an output from the unit corresponding to the control data output in step S250502 is input as a connection weight vector wji. In step S250505, the connection weight vector wji input from the unit in step S250504 is output to the outside.

  Through the above processing from step L250501 to step L250507, first, the coupling weight vector is output in order from each unit (step S250502), and then the coupling weight vector output by each unit in sequence according to step S250502 is input in order. (Step S250504). Next, the connection weight vector input in order from each unit in step S250504 is sequentially output to the outside as the calculation result of the self-organizing map (step S250505).

  The parallel computing device when applied to the self-organizing map according to the second basic configuration described above with reference to FIGS. 22 to 26 is the parallel computing device when adapted to the self-organizing map according to the first basic configuration. Similarly to the above, it is possible to execute the calculation of the self-organizing map and output the execution result to the outside.

  In addition, the parallel computing device when applied to the self-organizing map according to the second basic configuration is simpler than the parallel computing device when adapted to the self-organizing map according to the first basic configuration. There is an effect that there is. In addition, there is an effect that it is easy to change the design of the values of various variables in the self-organizing map, such as changing the number of dimensions of the input vector, learning speed, and neighborhood distance.

<Separation configuration>
Next, with reference to FIG. 27, the configuration of the parallel arithmetic device when the units and the distribution control device 8 are separated from each other in the parallel arithmetic device according to the second basic configuration of FIG. 14 will be described.

  The parallel arithmetic device shown in FIG. 27 has a connecting portion 9A and a connecting portion 9B that can be connected to each other with respect to the parallel arithmetic device shown in FIG. All the units AU1 to AU10 and the distribution control device 8 are connected by the unit output bus 2 and the unit input bus 3 through the connection 9A and the connection 9B. For example, the connecting portion 9A and the connecting portion 9B are desirably members that fit together.

  In addition, as shown in FIG. 27, all of the units AU1 to AU10 and the connection portion 9B are made into one device (device B) by the connection portion 9A and the connection portion 9B that can be connected to each other, and the distribution control device 8 and the connecting portion 9A are set as one device (device A).

  In this case, for example, the connecting portion 9A and the connecting portion 9B may be a female and male USB (Universal Serial Bus) (registered trademark) connector, the device B may be a USB device, and the device A may be a personal computer. . As described above, by configuring the unit and the distribution control device 8 as separate devices and using the distribution control device 8 as a personal computer, the configuration of the entire parallel arithmetic device becomes simpler and can be manufactured at low cost.

  Further, by making the distribution control device 8 a personal computer, it becomes possible to create a control procedure more easily by computer programming a predetermined control procedure and the like. Further, since the personal computer which is the distribution control device 8 also has a keyboard as input means and a monitor as display means or output means, the user can use the parallel arithmetic device more easily.

  Note that the connection unit 9A and the connection unit 9B are not limited to USB connectors, and may be connectors of other standards or communication network connectors. In the case of a communication network connector, the connection between the unit AU1 to the unit AU10 and the distribution control device 8 may be a communication connection using a communication protocol via the connection unit 9A and the connection unit 9B. The connection unit 9A and the connection unit 9B may control the communication protocol.

  Further, the device B may be configured as an expansion base device, and this expansion base device may be connected to an expansion slot of the distribution control device 8 which is a personal computer. For example, the device B configured as the expansion base device may be a board conforming to a PCI (Peripheral Component Interconnect) standard or a PCI-Express standard bus, for example. In this case, the personal computer as the device A may have an expansion slot conforming to, for example, the PCI standard or the PCI-Express standard as the connection unit 9A.

Each of the storage units having the configuration shown in the drawings of the above embodiment is a non-volatile memory such as a hard disk device, a magneto-optical disk device, or a flash memory, or a storage medium that can only be read such as a CR-ROM. , A volatile memory such as RAM (Random Access Memory), or a combination thereof.
The unit calculation unit in FIG. 15 in FIG. 2 or the distribution control unit 8 in FIG. 16 may be realized by dedicated hardware, or by a memory and a microprocessor. May be.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes a design and the like within the scope not departing from the gist of the present invention.

It is a block diagram which shows the structure of the parallel arithmetic unit by one Embodiment of this invention. It is a block diagram which shows the structure of the unit of the parallel arithmetic unit by one Embodiment of this invention. It is a block diagram which shows the structure of the unit of FIG. 2 in a hierarchical neural network. FIG. 4 is a first flowchart showing the operation of the unit of FIG. 3. It is a 2nd flowchart figure which shows operation | movement of the unit of FIG. It is explanatory drawing explaining a neural network. FIG. 3 is a first block diagram showing a configuration of a unit of FIG. 2 in a self-organizing map. It is a 2nd block diagram which shows the structure of the unit of FIG. 2 in a self-organization map. It is explanatory drawing which shows the data structure in a self-organization map. FIG. 9 is a first flowchart showing the operation of the unit of FIGS. 7 and 8. It is a 2nd flowchart figure which shows operation | movement of the unit of FIG. 7 and FIG. It is a block diagram which shows the structure of the parallel arithmetic unit which has a redundant circuit. It is a block diagram which shows the structure of the unit of FIG. It is a block diagram which shows the structure of the parallel arithmetic unit by a 2nd basic composition. It is a block diagram which shows the structure of the unit of FIG. It is a block diagram which shows the structure of the distribution control part 8 of FIG. It is a block diagram which shows the structure of the unit of FIG. 14 in a hierarchical neural network. 15 is a protocol table in the hierarchical neural network of FIG. It is a flowchart explaining operation | movement of each unit of FIG. It is the flowchart 1 explaining operation | movement of the distribution control part 8 of FIG. 15 is a flowchart 2 for explaining the operation of the distribution control unit 8 of FIG. 14. It is a block diagram which shows the structure of the unit of FIG. 14 in a self-organization map. It is a protocol table in the self-organization map of FIG. It is a flowchart explaining operation | movement of each unit of FIG. 23 is a flowchart 1 illustrating an operation of the distribution control unit 8 in FIG. 23 is a flowchart 2 illustrating the operation of the distribution control unit 8 in FIG. It is a block diagram which shows the structure of a parallel arithmetic unit at the time of setting the parallel arithmetic unit of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Daisy chain control bus 2 Unit output bus 3 Unit input bus 4 Amplifier 5 Trigger input line 6 1st redundant daisy chain control bus 7 2nd redundant daisy chain control bus 11, 101, 201, 11A, 101A, 201A Data input Unit 12, A12, B12, 102, 202 Token input unit 13, A13, B13, 103, 203 Token output unit 14, 104, 204, 14A, 104A, 204A Data output unit 16, 16A Unit output storage unit 17 Token output unit Selection unit 18 Token selection information storage unit 33 Control data output unit 101A, 201A Data input unit 104A, 204A Data output unit 150B, 250B Function number extraction unit 151B, 251B Function number storage unit 152B, 252B Land extraction unit 153B, 253B General-purpose buffer storage unit 111, 111A Product sum calculation primary value calculation unit 112, 112A Unit output calculation unit 113 Layer information calculation unit 114 Counter 130, 231, 231A Weight storage unit 131 Product sum calculation primary value storage unit 132 function storage unit 133 layer information storage unit 134 unit number storage unit 135, 135A unit output storage unit 136 data output flag storage unit 137 unit identification information storage unit 138B pre-stage unit output storage unit 210 data extraction unit 211 network coordinate extraction unit 212 norm Extraction unit 213 Counter 214, 214A Norm calculation unit 215, 215A Neighborhood determination unit 216, 216A Weight update unit 217 Norm comparison unit 218 Input norm determination unit 219 Network coordinate selection unit 220 Norm selection unit 221 Data composition output unit 222 Data input determination unit 223 Token output flag determination unit 230, 230A Input data storage unit 232, 232A Distance storage unit 233, 233A Input network coordinate storage unit 234, 234A Unit network coordinate storage unit 235, 235A Learning speed Storage unit 236, 236A Unit norm storage unit 237 Input norm storage unit 238 Token output flag storage unit 250 Input norm unit 251 Selection output unit L1 Input layer L2 Hidden layer L3 Output layer P1 Output terminal P2 Input terminal P3 Trigger input terminal U1, U2 , U3, U4, U5, U6, U7, U8, U9, U10, AU1, AU2, AU3, AU4, AU5, AU6, AU7, AU8, AU9, AU10 units

Claims (24)

Parallel computing device
A plurality of units each identified by a unique unit number that is a predetermined identification number;
One of the plurality of units selected from the input value and the plurality of units when an output value output from any one of the plurality of units is input via the unit output bus. A distribution control unit that outputs control data including a selection unit number that is an identification number for selecting one unit to each of the plurality of units via a unit input bus;
Have
The unit is
A data input unit for inputting control data from the distribution control unit via the unit input bus;
A unit number coincidence determining unit that determines whether or not the selected unit number included in the control data input to the data input unit matches the unique unit number;
A unit calculation unit for calculating by a predetermined calculation method for each unit based on an input value included in the control data input to the data input unit;
When the result determined by the unit number coincidence determination unit is coincident, the operation result, which is the result calculated by the unit operation unit, is output as the output value to the distribution control unit via the unit output bus. A data output section;
A parallel computing device comprising: The distribution controller is
The control data includes a control code for selecting a calculation method of the unit calculation unit,
The unit is
A control code extraction unit that extracts the control code from the control data input to the data input unit;
The unit calculation unit is
Based on the control data extracted by the control code extraction unit and the result determined by the unit number match determination unit, the calculation is performed, or the unit control unit is configured to execute the calculation.
The parallel arithmetic apparatus according to claim 1, wherein: The distribution control unit
A unit output value input unit for inputting an output value output from any one of the plurality of units via the unit output bus;
Based on the output value input to the unit output value input unit, a control data generation unit that generates the control data according to a predetermined processing procedure;
A control data output unit that outputs the control data generated by the control data generation unit to each of the plurality of units via the unit input bus;
The parallel arithmetic device according to claim 1, wherein the parallel arithmetic device is provided. The unit is
A unit output storage unit for storing the calculation result;
The unit calculation unit is
The calculation result is stored in the unit output storage unit,
The data output unit is
When the result determined by the unit number coincidence determination unit is coincident, the calculation result is read from the unit output storage unit, and the read calculation result is used as the output value as the output value to the distribution control unit. Output via
The parallel arithmetic apparatus according to claim 1, wherein The unit is
A unique unit number storage unit in which the unique unit number is stored in advance;
A selection unit number extraction unit that extracts the selection unit number from the control data input to the data input unit;
An input value extraction unit for extracting the input value from the control data input to the data input unit;
Have
The unit number match determination unit
Determining whether the selected unit number extracted by the selected unit number extraction unit and the unique unit number read from the unique unit number storage unit match;
The unit calculation unit is
Based on the input value extracted by the input value extraction unit, the calculation is performed by a predetermined calculation method for each unit,
The parallel arithmetic device according to claim 1, wherein When the parallel computing device is a parallel computing device that computes a hierarchical neural network,
The unit calculation unit is
Of the layers of the hierarchical neural network, a layer information storage unit storing layer information for identifying a layer to be output;
A function storage unit in which the output function of the unit and the layer information are associated and stored in advance;
A weight storage unit in which a predetermined coupling load multiplied by an output value from the unit of the output layer identified by the layer information is stored in association with the layer information;
A product-sum calculation primary value storage unit for storing product-sum calculation primary value information obtained by multiplying an input value from the output unit and a combined load;
In accordance with the selection of the unit control unit, the connection load corresponding to the layer information read from the layer information storage unit is read from the weight storage unit, and the read connection load and the input value input to the data input unit A product-sum calculation primary value calculation unit that adds and stores the multiplied value to the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit;
According to the selection of the unit control unit, the product sum calculation primary value information is read from the product sum calculation primary value storage unit, and the output function corresponding to the layer information read from the layer information storage unit is read from the function storage unit. A unit output calculation unit that calculates an output value by substituting the read product-sum calculation primary value information into the read output function, and stores the calculated output value in the unit output storage unit;
The parallel processing device according to claim 4, wherein: The unit arithmetic unit has a preceding unit number storage unit in which distribution unit identification information indicating which unit is outputting as an output unit among units belonging to the output layer is stored,
In the weight storage unit, a predetermined combined load multiplied by an output value from the unit of the output layer identified by the layer information and the distribution unit identification information, the layer information and the distribution unit identification information, Are stored in association,
The product-sum calculation primary value calculation unit reads the combined load corresponding to the distribution unit identification information read from the preceding unit number storage unit and the layer information read from the layer information storage unit from the weight storage unit, and reads the read The combined weight is multiplied by the input value input to the data input unit, and the multiplied value is added to and stored in the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit. ,
The parallel arithmetic apparatus according to claim 6. When the control code included in the control data input to the data input unit indicates to initialize the product-sum calculation primary value,
The unit controller
Initializing the product-sum calculation primary value information stored in the product-sum calculation primary value storage unit to 0;
The parallel arithmetic apparatus according to claim 6 or 7, characterized in that The control code included in the control data input to the data input unit indicates that the product sum calculation primary value is calculated and added and stored, and the unit of the unit in which the control data outputs the output value and the output value When including the distribution unit identification information that is identification information,
The unit controller
Storing the delivery unit identification information in the preceding unit number storage unit;
Causing the product-sum calculation primary value calculation unit to execute processing based on the output value;
9. The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus is characterized by that. When the control code included in the control data input to the data input unit indicates that the unit output calculation is performed,
The unit controller
Causing the unit output calculation unit to execute processing;
The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus according to claim 6. When the control code included in the control data input to the data input unit indicates that an output value is output as an active unit, and the control data includes a selected unit number,
The unit controller
When the result determined by the unit number coincidence determining unit is a match, the calculation result is read from the unit output storage unit, and the read calculation result is output as the output value to the distribution control unit and the data output unit. Output via the unit output bus,
The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus is characterized by that. When the control data input to the data input unit includes distribution layer information,
The unit controller
Storing the distribution layer information included in the control data input to the data input unit in the layer information storage unit;
12. The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus is characterized in that The distribution control unit
A control code indicating that the product-sum calculation primary value is initialized, the output value and the delivery unit identification information, and a control code indicating that the product-sum calculation primary value is calculated and added and stored; A control code indicating that an operation is executed, a control code indicating that an output value is output as the active unit, or control data including control data including the distribution layer information is stored in the number of layers of the hierarchical neural network. And in a predetermined order according to the number of nodes in the hierarchy,
13. The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus is characterized in that When the parallel arithmetic device is a parallel arithmetic device that calculates a self-organizing map, each of the plurality of units is identified by a unit network coordinate that is a network coordinate of the unit instead of the unique unit number,
The unit calculation unit is
A unit network coordinate storage unit in which the unit network coordinates are stored in advance;
A weight storage unit in which the unit weight vector is stored;
A unit norm storage unit for storing a unit norm which is a norm of the unit;
When control data including a control code indicating calculation of a norm and an input vector is input from the unit control unit to the data input unit, the input data vector input to the data input unit and the weight storage A norm calculation unit that calculates a norm with the connection weight vector read from the unit, and stores the calculated norm as the unit norm in the unit norm storage unit;
Have
When control data including a control code indicating that a norm is output and input network coordinates are input from the unit control unit to the data input unit,
The unit controller
If the input network coordinates input and the unit network coordinates read from the unit network coordinate storage unit match, the unit norm read from the unit norm storage unit is output via the data output unit,
The parallel arithmetic apparatus according to claim 4 or 5, wherein When the control code included in the control data input to the data input unit indicates that a weight is set and includes input network coordinates and an input coupling weight vector,
The unit control unit updates the value of the input network coordinates stored in the input network coordinate storage unit to the input network coordinates included in the received control data,
The unit number match determination unit determines whether or not the updated input network coordinates match the unit network coordinates read from the unit network coordinate storage unit,
When the determination results of the unit number match determination unit match, the unit control unit updates the connection weight vector stored in the weight storage unit to the input connection weight vector included in the received control data.
15. The parallel arithmetic apparatus according to claim 14, wherein When the control code included in the control data input to the data input unit indicates that the learning speed and the proximity distance are set, and includes the input learning speed and the input proximity distance,
The unit control unit updates the learning speed stored in the learning speed storage unit to the input learning speed included in the received control data, and includes the reference distance stored in the distance storage unit in the received control data Update to the input neighborhood distance
The parallel arithmetic apparatus according to claim 14 or 15, wherein When the control code included in the control data input to the data input unit indicates that the number of dimensions of the real vector is set and includes the number of input dimensions N,
The unit control unit updates the number of dimensions stored in the real vector dimension number storage unit to the number of input dimensions included in the received control data.
The parallel arithmetic apparatus according to claim 14, wherein the parallel arithmetic apparatus is characterized by that. When the control code included in the control data input to the data input unit indicates that the norm is calculated, and includes an input vector,
The unit control unit updates the input data vector stored in the input data storage unit to the input vector,
The norm calculation unit calculates a norm between the updated input vector and the combined weight vector read from the weight storage unit, and stores the calculated norm as a unit norm in the unit norm storage unit.
The parallel arithmetic apparatus according to claim 14, wherein The control code included in the control data input to the data input unit indicates that the norm is output. When the control code includes input network coordinates, the unit control unit is stored in the input network coordinate storage unit. Update the input network coordinate value to the input network coordinate included in the received control data,
The unit number match determination unit determines whether or not the updated input network coordinates match the unit network coordinates read from the unit network coordinate storage unit,
When the determination results of the unit number match determination unit match, the unit control unit reads the unit norm read from the unit norm storage unit via the data output unit and the unit output bus. Output to
19. The parallel arithmetic device according to claim 14, wherein the parallel arithmetic device is characterized by that. When the control code included in the control data input to the data input unit indicates learning and includes input network coordinates,
The unit control unit updates the input network coordinates stored in the input network coordinate storage unit to the input network coordinates included in the control data,
The computing unit is
When the control code included in the control data input to the data input unit indicates learning, the unit network coordinates read from the unit network coordinate storage unit and the input network coordinates read from the input network coordinate storage unit And the unit network coordinates read from the unit network coordinate storage unit is determined to be the input network by determining whether the calculated distance is equal to or less than the distance read from the distance storage unit. A proximity determining unit that determines whether or not the input network coordinates are read from the coordinate storage unit;
If the result of the determination by the proximity determination unit is a neighborhood, the difference between the input data vector read from the input data storage unit and the combined weight vector read from the weight storage unit is calculated, and the calculated difference is A weight update unit that adds a value obtained by multiplying the learning speed coefficient read from the learning speed storage unit to the connection weight vector stored in the weight storage unit, and updates the connection weight vector stored in the weight storage unit; ,
20. The parallel arithmetic device according to claim 14, wherein the parallel arithmetic device comprises: When the control code included in the control data input to the data input unit indicates that the learning result is output and includes the input network coordinates,
The unit control unit updates the value of the input network coordinates stored in the input network coordinate storage unit to the input network coordinates included in the received control data;
The unit number match determination unit determines whether the updated input network coordinates match the unit network coordinates read from the unit network coordinate storage unit,
When the determination results of the unit number coincidence determination unit match, the unit control unit outputs the connection weight vector read from the weight storage unit via the data output unit and the unit output bus to the distribution control unit. Output to
21. The parallel arithmetic apparatus according to claim 14, wherein the parallel arithmetic apparatus is characterized by that. The distribution control unit
A control code indicating setting the weight, a control code indicating setting the learning speed and the neighborhood distance, a control code indicating setting the number of dimensions of a real vector, a control code indicating calculating a norm, Control data including a control code indicating that a norm is output, a control code indicating that learning is performed, or a control code that indicates that a learning result is output is determined in advance according to the number of units and the number of times of learning. Output in the order given,
13. The parallel arithmetic apparatus according to claim 6, wherein the parallel arithmetic apparatus is characterized in that The unit output bus and the unit output bus that connect the plurality of units and the distribution control unit are separated by a first connection unit and a second connection unit that can be connected to each other,
A first device including the plurality of units and the first connection unit;
A second device including the distribution control unit and the second connection unit;
The parallel arithmetic apparatus according to any one of claims 1 to 22, wherein A plurality of units each identified by a unique unit number that is a predetermined identification number;
One of the plurality of units selected from the input value and the plurality of units when an output value output from any one of the plurality of units is input via the unit output bus. A distribution control unit that outputs control data including a selection unit number that is an identification number for selecting one unit to each of the plurality of units via a unit input bus;
A parallel operation used in a parallel operation device having
The unit is
Input control data from the distribution control unit via the unit input bus,
Determining whether the selected unit number included in the control data input to the data input unit matches the unique unit number;
Based on the input value included in the input control data, calculated by a predetermined calculation method for each unit,
When the determined result is coincident, the calculation result which is the calculation result is output as the output value to the distribution control unit via the unit output bus.
A parallel operation method characterized by that.

Images