JPH0373056A - Data processor and rationalizing method for the same - Google Patents

Abstract

PURPOSE:To prevent a drop into a local minimum in a learning process by providing plural neurons to output data corresponding to the result of comparison between the total sum obtd. by multiplying prescribed weight to input data and a threshold value, and possessing distribution that the threshold value of the neuron is changed like a step in an initial state. CONSTITUTION:Plural neural layers NL, for which plural neurons N are provided parallelly, are provided and one neural layer is constituted so that the output of a certain neural layer can be the input of the neural layer in the next step. Threshold values theta of the neurons in the same neural layer are made same and the threshold value theta is increased toward the neural layer in the rear step and changed like a step. Normally in the learning process, only the neuron close to an input side is ignited in the beginning and the range of the ignition is gradually spread to the neurons in the rear steps. In such a process, a process for minimizing the threshold value is led in, a lot of neurons are related to data processing at a time point that the neural layer in the final step is ignited. Thus, the drop into the local minimum can be prevented.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention provides a data processing system that is provided with a plurality of neurons that output data according to a comparison result between the sum of input data multiplied by a predetermined weight and a threshold value. The present invention relates to a processing device and a method for optimizing the weights in the data processing device.

[Conventional technology]

During the learning process of this type of data processing device, a certain degree of proper input-output correlation occurs, and when that correlation becomes stronger, the data processing device may be unable to escape from that state and optimization may become impossible. .

This is expressed as a fall to a local minimum, for example, the Polzmann machine (David H, Ackley,
Geoffrey E, Hinton+ and
Terrence J, Sejnowski: Al
learning a1gorithn+ for Bo
lzn+an machines: Cognitive
5cfence 9+ 1989+ 147-169
), the energy is temporarily jumped to a predetermined high level in order to escape from the energy equation of the neural network and reach the global local minimum. However, the optimum correlation between input and output does not necessarily match the state of the global and local energy. In addition, the depth of the local minimum is generally unknown, and there is no guarantee that the energy level will definitely escape from the local minimum when the energy level is increased by a predetermined value, and that the global local minimum of energy can be reached. There's not even a guarantee. Furthermore, a temporary jump in energy itself is inconsistent with the functioning of biological systems.

[Problems to be Solved by the Invention] The present invention has been devised to solve the above-mentioned conventional problems, and provides a data processing device and its optimization that can reliably prevent a drop to a local minimum in the learning process. The purpose is to provide a method.

[Means to solve the problem]

The data processing device according to the present invention gives a smooth gradient bias to the threshold distribution of neurons in an initial state,
Further, the optimization method according to the present invention is such that, while giving a constant input to the data processing device, the threshold value of a neuron that has produced a significant output at a certain point in time is raised to the maximum value, and then reduced to the value at the said point in time. The threshold value is forcibly changed, and the weight is adjusted during the period of this forced change of the threshold value.

[Effect]

According to the method for optimizing a data processing device according to the present invention, it is possible to reliably escape from a local minimum value using a mechanism that conforms to the function of a biological system. Furthermore, the data processing device according to the present invention can perform effective optimization and distortion using the optimization method.

〔Example〕

The present invention will be explained below with reference to illustrated embodiments.

Here, as an indicator showing the effectiveness of learning, the following standard - maturity T
Let us define P.

(Left below) However, where n is the number of two-input events i, j: number of input events Pi, j: degree of match between the neural network output due to the i-th input event and the neural network output due to the j-th input event. (Number of matching bits) The above-mentioned standard-quality TP is an index of the discrimination ability of input events in the neural network, and the lower the TP, the higher the discrimination ability. In a neural network that has not been trained at all, TP = 100%.
, and TP monotonically decreases with subsequent learning.

(Figure 1) However, if the learning method is inappropriate, it may actually impede the learning effect, and as shown in Figure 2, the TP that once decreased may rise again. .

When a drop to the local minimum occurs as described above, TP becomes saturated at a high value as shown in FIG.

This local drop to the local minimum is unrelated to the energy equation (1) above, and may be caused by, for example, bias in the weight of the features of a certain event, and a certain feature is emphasized, resulting in confusion of multiple events. This can be thought of as the result of learning to affirm .

Expressing this schematically, as shown in Figure 4, because the influence of a certain input bit 4 is strong, the firing distribution of neurons becomes biased, and this state can be considered as a state that continues to be strengthened. . In this figure, neurons that have fired are indicated by black circles.

In order to escape from this state, other input bits 11 to I
It is necessary to strengthen neuronal involvement in s, Is.

Here, the inventors focused on the absolute refractory period and relative refractory period in neural networks of biological systems.

These refractory periods are said to contribute to suppressing and controlling the propagation of sound waves in the neural networks of biological systems. (Toshi Seri: Mathematics of Neural Networks - Information Processing Style of the Brain -; Sangyo Tosho) On the other hand, the inventors believe that these refractory periods temporarily balance the effects of each feature of the input data. Considering this as a changing factor, we changed the threshold of a neuron firing in response to a certain input as shown in Figure 5, giving the same effect as the absolute refractory period and the relative refractory period.

That is, the threshold value θ of a neuron fired by association at a certain time point to in the learning process is then increased to the maximum value θmax (for example, infinity) for a certain period of time to. This corresponds to the absolute refractory period. Even during this absolute refractory period, learning continues and changes the weights of neurons. The neuron to which the weight change is applied at this time is a neuron that did not fire in response to that input before to. After the period ta has elapsed, the threshold value is gradually decreased to the value θ0 at time 10. This period of gradual decline corresponds to the relative refractory period, and this period is designated as tr.

In the absolute refractory period, only neurons that did not fire at time 10 are weighted, and learning is performed on a biased neuron distribution, but in the relative refractory period,
The neurons that fired at that point also gradually start firing, and balanced corrective learning is gradually performed for the entire neural network.

As a result, an escape from the so-called local minimum is realized, and learning is performed to bring the standard-political change to the optimal value (global local minimum).

As shown in FIG. 6, by giving the refractory period a plurality of cycles, escape from the local minimum becomes more reliable. However, if the number of refractory period cycles is too large, the convergence of learning will be hindered, and even if the refractory period cycles are given at the end of learning, the convergence of learning will be inhibited.

That is, the refractory cycle needs to be given an appropriate number of times at a relatively early stage of learning.

It is also necessary to consider optimization of the absolute refractory period ta, the relative refractory period tr, and the balance between the two.

Next, a data processing apparatus in which the above optimization method is effectively executed will be described.

In FIG. 7, the data processing device has a plurality of neural layers NL each having a plurality of neurons N arranged in parallel, and the neural layers are such that the output of one neural layer becomes the input of the next neural layer. It is composed of

In such a configuration, the topology of the neural network can be defined, and the coordinates of each neuron can be specified as shown in FIG.

Here, the X-axis is taken in the direction from the input side to the output side, and the Y-axis is taken in the width direction of each neural layer.

Then, the threshold distribution of neurons in the initial state of the data processing device is set as shown in FIG. In other words, the threshold value θ of neurons in the neural layer is set to be the same, and the threshold value θ is set higher as the neural layer becomes later. The gradient of increase in the threshold value is assumed to change stepwise.

That is, the threshold value has, for example, the same magnitude for three consecutive neural layers and increases for every third neural layer. The rate of increase may be uniform over the entire neural network, but it may be larger at later stages as shown in FIG.

In the normal learning process, the weight of synapses with significant synaptic input is increased in response to correct input, but in the threshold distribution shown in Figure 8, initially only neurons closer to the input side fire, and gradually later neurons fire. Neurons have a wider firing range.

In this process, introducing the threshold maximization process described above has the effect of increasing the number of firing neurons for a given input in each neural layer, and by the time the final neural layer starts firing, many Neurons become involved in data processing and do not converge on a biased firing pattern. Therefore, it is possible to prevent a drop to the local minimum.

FIG. 9 shows a second embodiment, in which the initial threshold distribution is such that the threshold of a neuron at the center of the data processing device is increased, and the threshold is lowered stepwise as the distance from this central neuron increases. In such a data processing device, a firing pattern occurs around the base of the threshold value in the initial stage of learning. However, when maximizing the threshold as described above, there are neurons with relatively low thresholds adjacent to the firing pattern, so
The distribution of firing neurons gradually widens toward the top of the threshold. As a result, many neurons become involved in data processing at the end of learning, and the firing pattern does not converge to a biased one.

Therefore, it is possible to prevent a drop to the local minimum.

FIG. 10 shows the third embodiment, and the initial threshold distribution is as follows:
The threshold value of the neuron at the median value of the Y coordinate is set to the lowest value, and the threshold value is increased in a step-like gradient toward the maximum value and the minimum value of the Y coordinate. In such a data processing device, at the beginning of learning, Y
A firing pattern occurs that penetrates around the coordinate median value. However, when maximizing the threshold as described above, there are neurons with relatively low thresholds adjacent to the firing pattern, so the distribution of firing neurons expands toward both the maximum and minimum Y coordinates. go. As a result, at the end of the study,
Many neurons will be involved in data processing,
It does not converge to a biased firing pattern. Therefore, falling to the local minimum can be prevented.

FIG. 11 shows a fourth embodiment. This embodiment differs from the first embodiment shown in FIG. 8 in that the threshold values of neurons in all neural layers are the same at the same Y-coordinate, whereas in the same neural layer, the more the Y-coordinate increases, the more neurons The threshold value θ of is increased. The gradient of increase in this threshold value is stepwise. In the threshold distribution shown in FIG. I1, initially only neurons with large Y coordinates fire, and the firing range gradually expands to neurons with small Y coordinates. Here, when the threshold is maximized as described above, there are neurons with relatively low thresholds adjacent to the firing pattern, so the distribution of firing neurons spreads even more smoothly. As a result, many neurons will be involved in data processing at the end of learning, preventing convergence to a biased firing pattern and preventing a drop to a local minimum.

FIG. 12 shows a fifth embodiment, in which the thresholds of neurons in all neural layers are the same at the same Y-coordinate, while at a particular Y-coordinate (e.g. central ) The threshold value θ at M takes the maximum value, and the threshold value θ decreases stepwise as the Y coordinate becomes further away from this value. In the threshold distribution shown in FIG. 12, neurons on the maximum and minimum Y coordinate side initially fire, and the range of firing gradually expands to neurons on the YFi mark M side. Here, when the threshold is maximized as described above, there are neurons with relatively low thresholds adjacent to the firing pattern, so the distribution of firing neurons spreads even more smoothly, preventing a drop to the local opportunism. be done.

FIG. 13 shows a sixth embodiment. In this example, the threshold value θ of the neuron decreases as the X and Y coordinates become smaller, and increases in a stepwise manner as the X and Y coordinates become larger. In this threshold distribution, initially the X coordinate and Y
Neurons with smaller coordinates fire, and the firing range gradually expands to neurons with larger X and Y coordinates. If the threshold is maximized as described above, there will be neurons with relatively low thresholds adjacent to the firing pattern, so the distribution of firing neurons will spread even more smoothly, and a drop to the local minimum will be prevented. Ru.

FIG. 14 shows a seventh embodiment, in which the initial threshold distribution is such that the threshold of the central neuron is the minimum value, and the threshold increases stepwise as the distance from the central neuron increases. In this example, a firing pattern is generated from the central neuron at the beginning of learning. Here, when the threshold is maximized as described above, there are neurons with relatively low thresholds adjacent to the firing pattern, so the distribution of firing neurons gradually expands toward surrounding neurons. As a result, many neurons become involved in data processing at the end of learning, preventing convergence to a biased firing pattern and preventing a drop to a local minimum.

FIG. 15 shows the eighth embodiment, where Y is the initial threshold distribution.
Regarding the coordinates, for example, the center has a high threshold value, and regarding the X coordinate, for example, the center has a low threshold value. That is, the threshold distribution has a saddle shape, and the change in the distribution is step-like. This embodiment also provides the same effects as those of the above embodiments.

[Effects of the Invention] As described above, the method for optimizing a data processing device according to the present invention has an excellent effect of reliably escaping the local minimum.

Furthermore, according to the data processing device of the present invention, since the initial threshold distribution is biased with a step-like gradient, the firing pattern spreads in a way that eliminates this bias during the learning process, and as a result, the bias is convergence to a certain firing pattern is prevented.

[Brief explanation of drawings]

Figure 1 is a graph showing the change from standard to political change when proper learning is carried out. Figure 2 is a graph showing the change from standard to political change when inappropriate learning is carried out. Figure 3 is a graph showing the change from standard to political change when inappropriate learning is carried out. A graph showing the change from standard to political change when the process falls to local control, Figure 4 is a conceptual diagram showing an example of the firing state of a neural network that falls to local control, and Figure 5 is the book. FIG. 6 is a graph showing threshold changes according to one embodiment of the invention method; FIG. 7 is a conceptual diagram showing a first embodiment of the data processing device according to the invention; FIG. FIG. 8 is a graph showing the initial threshold distribution of the same example, FIG. 9 is a graph showing the initial threshold distribution of the second example, FIG. 10 is a graph showing the initial threshold distribution of the third example, and FIG. 11 is a graph showing the initial threshold distribution of the third example. FIG. 12 is a graph showing the initial threshold distribution of the fifth embodiment. FIG. 13 is a graph showing the initial threshold distribution of the sixth embodiment. FIG. 14 is the graph showing the initial threshold distribution of the fifth embodiment. Graph showing the initial threshold distribution of the embodiment. FIG. 15 is a graph showing the initial threshold distribution of the eighth embodiment. Figure 1 Number of learning times

Claims (2)

[Claims] (1) In a data processing device equipped with multiple neurons that output data according to the comparison result between the sum of input data multiplied by a predetermined weight and a threshold value, the threshold values of the neurons change in a stepwise manner in the initial state. A data processing device characterized by having a distribution. (2) Data processing in which multiple neurons are provided that output data according to the comparison result of the sum of input data multiplied by a predetermined weight and a threshold value, and the threshold value of the neuron has a distribution that changes stepwise in the initial state. This is a device optimization method in which the threshold of a neuron that has produced a significant output at a certain point is forcibly changed so that it temporarily increases to the maximum value and then decreases to the value at that point. A method for optimizing a data processing device, comprising: modifying the weights based on output evaluation results while giving constant input data within a period of forced change.