JP7118859B2 - Information processing system and information processing method - Google Patents

Description

The present invention relates to an information processing system and an information processing method for realizing time-series data processing by machine learning using a physical system.

In the brain of a living body, a large number of neurons form a network through synaptic connections, and these are considered to be responsible for information processing. A system that imitates such a structure on a computer is called a neural network.

Neural network models are roughly classified into two types. A model with a unidirectional signal flow and a non-looping structure is called a feedforward neural network. On the other hand, a model that has self-feedback inside and has a structure in which signals propagate in both directions is called a recurrent neural network.

Neural networks acquire the ability to optimize and solve different tasks by varying the strength of connections between each neuron. Optimization methods are broadly classified into two. A method of learning by using correct data given in advance as input is called supervised learning, and a method of extracting the structure of data without using correct data is called unsupervised learning.

Reservoir computing is one form of realizing supervised learning by a recurrent neural network (see Non-Patent Document 1). Reservoir computing is performed by the following procedure. A neural network is composed of an input layer, an intermediate layer, and an output layer. Based on the output results obtained by inputting the training data, the strength of the connections from the input layer to the intermediate layer, from the intermediate layer to the intermediate layer, and from the intermediate layer to the output layer are optimized. do. In the case of a recurrent neural network, the learning algorithm becomes complicated, and a huge amount of calculation is required to optimize the connection strength. In contrast, in reservoir computing, the strength of the connection from the input layer to the middle layer and the strength of the connection from the middle layer to the middle layer are each fixed at a constant value, and the strength of the connection from the middle layer to the output layer is Only for In that case, the learning algorithm becomes simpler and the amount of calculation required for optimization can be reduced. In situations where learning takes place in this fashion, the intermediate layer is called a reservoir.

Here, in order to efficiently perform time-series data processing by reservoir computing, the reservoir must satisfy three characteristics (see Non-Patent Document 2).

The first feature is "nonlinearity". We show that the input signal can be nonlinearly transformed into a higher-dimensional state space. Such non-linearity can be achieved by increasing the number of neurons in the reservoir and increasing the number of recursive connections.

The second feature is the "short-term memory effect". We show that the state of the reservoir is affected by the most recent past input signal, but at the same time is not affected by the older past input signal. This property is necessary when performing time-series data processing such as speech recognition, in which the most recent past information is more important.

The third feature is "reproducibility". It shows that the same output signal is always obtained for the same input signal. Without reproducibility, learning cannot optimize the bond strength.

In order to realize a reservoir that satisfies the above three characteristics, there are a method using computer simulation and a method using a physical system. Higher-order nonlinearity is required for complex time-series data processing. When computer simulation is used, an enormous amount of calculation is required to obtain high-order nonlinearity. Therefore, attempts have been made to perform reservoir computing using physical systems.

Here, there is a technique that utilizes the nonlinearity and reproducibility of an optical intensity modulator and uses a loop structure using an electric circuit and an optical fiber. For example, the technique described in Non-Patent Literature 3 is configured with optical components that exhibit a simple response, so high reproducibility can be obtained. There is also disclosed a technique that utilizes the nonlinearity and reproducibility of a semiconductor laser and uses a feedback loop formed by a semiconductor laser and return light. For example, the technique described in Non-Patent Document 4 generates chaos when returning light is input to a semiconductor laser, so high nonlinearity can be obtained.

International Publication No. 2017/131081 JP-A-8-18139

H. Jaeger and H. Haas, Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication. Science 304, 78 (2004). L. Appeltant et al., Information processing using a single dynamical node as a complex system. Nature Communications 2, 468 (2011). Y. Paquot et al., Optoelectronic Reservoir Computing. Scientific Reports 2, 287 (2012). J. Nakayama et al., Laser dynamical reservoir computing with consistency: an approach of a chaos mask signal, Optics Express 24, 8679 (2016).

However, since it is generally difficult to control the stability of physical systems that exhibit complex responses, there is the problem that it is difficult to achieve both high nonlinearity and reproducibility.

An object of the present invention is to solve the above problems and to provide an information processing system and an information processing method for efficiently realizing time-series data processing by machine learning using a physical system.

To achieve the above object, an information processing system according to the present invention includes a signal conversion unit that holds at least part of a signal input in the past as an internal state, and nonlinearly converts the input signal in the internal state; A signal input unit that inputs the input signal to the signal conversion unit, a signal output unit that reads an output signal that is a signal converted by the signal conversion unit, and a linear combination and an optimum coupling strength of the read output signal. a calculation unit for calculating, wherein the signal conversion unit includes an optical amplifier that non-linearly converts the input signal, which is an optical signal, to the output signal, which is an optical signal.

ADVANTAGE OF THE INVENTION According to this invention, the information processing system which implement|achieves time series data processing by machine learning efficiently can be provided.

Problems, configurations, and effects other than those described above will be clarified by the following description of the mode for carrying out the invention.

1 is a diagram illustrating a configuration example of an information processing system according to Example 1 of the present invention; FIG. It is a figure which shows the structural example of an optical intensity modulator. It is a figure which shows the structural example of an optical amplifier. 4 is a flow chart showing an operation example of the calculation unit of the information processing system according to Example 1 of the present invention; It is a figure which shows the structural example of the information processing system which concerns on Example 2 of this invention. It is a figure which shows the simulation result by the information processing system which concerns on Example 2 of this invention. It is a figure which shows the structural example of the information processing system which concerns on Example 3 of this invention. It is a figure which shows the structural example of the computer which implement|achieves the computer of the information processing system which concerns on each Example of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description and drawings are examples for explaining the present invention, and are appropriately omitted and simplified for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

When there are a plurality of components having the same or similar functions, they may be described with the same reference numerals and different suffixes. However, if there is no need to distinguish between these multiple constituent elements, the subscripts may be omitted in the description.

Also, in the following description, there are cases where processing performed by executing a program will be described, but the program is executed by a processor (for example, CPU, GPU) to appropriately perform the specified processing using storage resources ( For example, a memory) and/or an interface device (for example, a communication port) are used, so the main body of processing may be a processor. Similarly, a main body of processing executed by executing a program may be a controller having a processor, a device, a system, a computer, or a node. The subject of the processing performed by executing the program may be an arithmetic unit, and may include a dedicated circuit (for example, FPGA or ASIC) that performs specific processing.

Programs may also be installed on a device, such as a computer, from a program source. The program source may be, for example, a program distribution server or a computer-readable storage medium. When the program source is a program distribution server, the program distribution server may include a processor and storage resources for storing the distribution target program, and the processor of the program distribution server may distribute the distribution target program to other computers. Also, in the following description, two or more programs may be implemented as one program, and one program may be implemented as two or more programs.

An information processing system of the present invention includes a signal converter. The signal converter is implemented by a physical system that includes optical amplifiers and acts as a reservoir. A signal input unit for inputting a signal to the signal conversion unit, a signal output unit for reading out the output signal that is the signal converted by the signal conversion unit, and a linear combination of the read output signals and calculating the optimal coupling strength of the neural network. It operates as a reservoir computer by being provided with a calculation unit. The part that realizes the input layer in reservoir computing is the signal input part, the part that realizes the intermediate layer, that is, the reservoir, is the signal conversion part, the part that realizes the output layer is the signal output part, and the part that learns the connection strength is the calculation part. Become.

"Reservoir computing" is one of the methods of supervised machine learning using recurrent neural networks. It is suitable for processing time-dependent information such as voice. Suitable for operation. In "reservoir computing", among the input layer, intermediate layer, and output layer that constitute the neural network, an intermediate layer that is a recurrent neural network with fixed weights is called a "reservoir." In "Reservoir Computing", the weights of the linear combinations from the input layer to the reservoir are fixed, the weights of the linear combinations within the reservoir are also fixed, and only the weights of the linear combinations from the reservoir to the output layer are updated by learning. .

Each embodiment and each modification described below can be combined in whole or in part without departing from the scope of the technical idea of the present invention.

(Configuration of information processing system of embodiment 1)
FIG. 1 is a diagram showing a configuration example of an information processing system according to Embodiment 1 of the present invention. In the information processing system 1S according to the first embodiment, the signal input section 1 is composed of a D/A converter 102, a light source 103, and an optical intensity modulator 104. FIG. Each part is connected by an electric wire 105 indicated by a dotted line arrow or an optical fiber 106 indicated by a solid line arrow. A computer (calculating unit) 101 includes a D/A converter 102 and an A/D converter 109, and is capable of outputting and measuring electrical signals. An electrical signal output from D/A converter 102 based on audio data is input to optical intensity modulator 104 . The light from the light source 103 is input to the optical intensity modulator 104 and undergoes intensity modulation based on an electrical signal. Although an electrical signal based on audio data is output from the D/A converter 102 and input to the optical intensity modulator 104, data other than audio data may be used.

(Configuration of intensity modulator)
FIG. 2 is a diagram showing a configuration example of an optical intensity modulator. An electro-optic crystal 201 is sealed in the light intensity modulator 104 . When an input optical signal 202 is input, it passes through an optical waveguide interferometer formed in the electro-optic crystal 201 and an output optical signal 203 is output. When an input electrical signal 204 is applied to the electro-optic crystal 201, the refractive index of the optical waveguide changes, the optical signals interfere and cancel each other, and the intensity of the output optical signal 203 changes. Due to the operation described above, the output optical signal 203 exhibits a nonlinear response to the input electrical signal 204 . The intensity-modulated light output from the optical intensity modulator 104 becomes an optical signal output from the signal input section.

Returning to the description of FIG. The signal converter 2 is configured by an optical amplifier 107 . The optical amplifier 107 receives an intensity-modulated optical signal output from the signal input section.

(Configuration of optical amplifier)
FIG. 3 is a diagram showing a configuration example of an optical amplifier. The optical amplifier 107 is an EDFA (Erbium Doped Optical Fiber Amplifier), in which an impurity-doped optical fiber 301, two wavelength division multiplexing optical couplers 302 and 303, and a pumping light source 304 are enclosed.

Pumping light 305 is output from the pumping light source 304 . An input optical signal 306 is multiplexed with pumping light 305 by a wavelength division multiplexing optical coupler 302 and input to an impurity doped optical fiber 301 . The excitation light 305 excites the energy of impurity ions in the impurity-doped optical fiber 301, and the input optical signal 306 causes the stimulated emission of the energy of the impurity ions. The remaining pumping light 307 is demultiplexed by the wavelength division multiplexing optical coupler 303, and the input optical signal 306 amplified by stimulated emission is output as an output optical signal 308. FIG.

Due to the operation described above, the output optical signal 308 exhibits a nonlinear response to the input optical signal 306 . Since the excited state transitions to the ground state in a constant time, it has a short-term memory effect. If the intensity of the input light and the internal energy distribution are the same, the light of the same intensity is output, so reproducibility can be obtained.

As described above, all three characteristics that a reservoir should have are achieved. The intensity-converted optical signal output from the optical amplifier 107 becomes the optical signal output from the signal converter.

The optical amplifier 107 is not limited to the EDFA, and may be a semiconductor optical amplifier that can expect a memory effect with a time width different from that of the EDFA.

Returning to the description of FIG. The signal output section 3 is composed of a photodetector 108 and an A/D converter 109 . The optical signal output from the signal converter 2 is input to the photodetector 108 and converted into an electrical signal. The electrical signal is input to the A/D converter 109 provided in the computer 101, and the computer 101 records the measured value.

The calculation unit is configured by the computer 101 . The computer 101 includes an arithmetic processing device such as a CPU (Central Processing Unit). FIG. 4 is a flow chart showing an operation example of the calculation unit of the information processing system according to the first embodiment of the present invention. First, the computer 101 inputs teacher data to the optical intensity modulator 104 through the D/A converter 102 (step S401). Next, the computer 101 operates the signal input unit 1, the signal conversion unit 2, and the signal output unit 3, and based on the measurement value obtained through the A/D converter 109, the coupling strength set inside the computer 101. (step S402).

Next, the computer 101 compares the output result obtained in step S402 and the answer of the teacher data, and determines whether they match (step S403). If they do not match (step S403: YES), computer 101 updates the coupling strength so as to reduce the error (step S404), and inputs the teacher data to optical intensity modulator 104 again. By repeating this loop of steps S401→S402→S403: NO→S404→S401 until the output result matches the answer of the teacher data, the coupling strength (linear weight) is optimized. Since optimization is performed only for the final output layer, optimization is completed in a short time compared to machine learning methods that optimize all connection strengths.

If the output result obtained by completing the optimization matches the answer of the teacher data (step S403: YES), the computer 101 inputs unknown data (step S405). When the output result obtained by completing the optimization matches the answer of the teacher data (step S403: YES), it means that the classification of the teacher data is completed. If the output result obtained by completing the optimization matches the answer of the teacher data, the computer 101 obtains an output result even for unknown data (step S406).

As described above, according to this embodiment, by utilizing the nonlinear dynamics of an optical amplifier, both high nonlinearity and reproducibility can be achieved in time series data processing by machine learning using physical systems. can be efficiently realized.

In the first embodiment, the optical amplifier 107 alone is used as the signal converter, but in the second embodiment, the optical amplifier is incorporated in a loop made up of an optical fiber, so that the nonlinearity and the short-term memory effect of the signal converter can be further reduced. You can control it freely.

FIG. 5 is a diagram showing a configuration example of an information processing system according to Embodiment 2 of the present invention. The information processing system 2S of the second embodiment differs from the information processing system 1S of the first embodiment in that the optical intensity modulator 104, the optical amplifier 107, the optical coupler 501, the optical isolator 502, and the optical filter 503 each include an optical fiber 106. It has a loop structure L2 configured by connecting with .

A part of the optical signal branched from the electro-optic coupler 201 is input to the photodetector 108 . The electrical signal output from the photodetector 108 is input to the A/D converter 109, and the computer 101 extracts the clock frequency. When the phase of the light in the loop structure L2 changes due to changes in temperature or the like, it appears as a change in the clock frequency. By controlling the clock frequency of the electrical signal output from the D/A converter 102 based on the extracted clock frequency, the optical signal generated in the loop structure L2 is stabilized. After that, by inputting teacher data and optimizing coupling coefficients in the same manner as in the first embodiment, it is possible to classify speech data and the like.

(Simulation result of Example 2)
FIG. 6 is a diagram showing simulation results by the information processing system according to the second embodiment of the present invention. An input electrical signal 601 to the reservoir that is output from the D/A converter 102 and an output electrical signal 602 from the reservoir that is input to the A/D converter 109 are shown. The input electrical signal 601 is a signal having a waveform with two amplitudes, a first relatively large amplitude and a second relatively small amplitude, while the output electrical signal 602 is compared to the input electrical signal 601. This resulted in a signal with a complex waveform. This is because the output electrical signal 602 is influenced by the past input electrical signal 601, and indicates that nonlinearity and short-term memory effect are exhibited. It was also confirmed that when the input electric signal 601 had the same waveform, the output electric signal 602 had the same waveform, and reproducibility was obtained.

As described above, according to the present embodiment, time-series data processing can be efficiently realized by utilizing the nonlinear dynamics of the loop structure of the optical amplifier and the optical fiber.

In the second embodiment, the clock extraction and clock frequency control by the computer 101 stabilize the optical signal. can be stabilized.

(Information processing system of embodiment 2)
FIG. 7 is a diagram showing a configuration example of an information processing system according to Embodiment 3 of the present invention. Compared to the information processing system 2S of the second embodiment, the information processing system 3S of the third embodiment has a feedback control system including a photodetector 108, a clock extractor 701, a phase shifter 702, and an optical phase modulator 703. have The reservoir in FIG. 7 is in a configuration called replay mode synchronization. A part of the optical signal branched from the optical coupler 501 is input to the photodetector 108 . A part of the electrical signal output from the photodetector 108 is input to the clock extractor 701 to extract the clock frequency. When the phase of the light in the loop structure L2 changes due to changes in temperature or the like, it appears as a change in the clock frequency. By inputting the electrical signal output from the clock extractor 701 to the optical phase modulator 703 through the phase shifter 702, the phase change occurring in the loop structure L2 is corrected and the optical signal occurring in the loop structure L2 is stabilized. . After that, by inputting teacher data and optimizing coupling coefficients in the same manner as in the second embodiment, it becomes possible to classify speech data and the like.

As described above, according to the present embodiment, time-series data processing can be efficiently realized by utilizing the nonlinear dynamics of the loop structure of the optical amplifier and the optical fiber and performing stabilization by regenerative mode locking. can.

FIG. 8 is a diagram showing a configuration example of a computer that implements the computer of the information processing system according to each embodiment of the present invention. A computer 5000 that realizes the computer 101 of the information processing systems 1S to 3S of Embodiments 1 to 3 includes a CPU (Central Processing Unit) 5300, a memory 5400 such as a RAM (Random Access Memory), an input device 5600 (for example, a keyboard, a mouse, touch panel, etc.) and an output device 5700 (eg, a video graphics card connected to an external display monitor) are interconnected through memory controller 5500 . In the computer 5000, a predetermined program is read from an external storage device 5800 such as an SSD or HDD via an I/O (Input/Output) controller 5200, and executed by cooperation of the CPU 5300 and the memory 5400. Computers 101 of information processing systems 1S to 3S are realized. Alternatively, the program for realizing computer 101 of information processing systems 1S-3S may be obtained from an external computer through communication via network interface 5100. FIG.

The present invention is not limited to the above-described embodiments, and includes various modifications. For example, the above-described embodiments have been described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the described configurations. It is also possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Moreover, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

1S, 2S, 3S... information processing system, 1... signal input unit, 2... signal conversion unit, 3... signal output unit, L2... loop structure, FC... feedback control system, 101... calculator (calculation unit), 102... D /A converter 103 light source 104 optical intensity modulator 105 electric wire 106 optical fiber 107 optical amplifier 108 photodetector 109 A/D converter 201 electro-optic crystal 202 Input optical signal 203 Output optical signal 204 Input electrical signal 301 Impurity-doped optical fiber 302 Wavelength division multiplexing optical coupler 303 Wavelength division multiplexing optical coupler 304 Pumping light source 305 Pumping Light 306 Input optical signal 307 Pumping light 308 Output optical signal 501 Optical coupler 502 Optical isolator 503 Optical filter 601 Input electrical signal to reservoir 602 Output from reservoir Electrical signal 701 Clock extractor 702 Phase shifter 703 Optical phase modulator.

Claims (3)

a signal conversion unit that holds at least part of a signal input in the past as an internal state and non-linearly converts the input signal in the internal state;
a signal input unit that inputs the input signal to the signal conversion unit;
a signal output unit that reads an output signal that is a signal converted by the signal conversion unit;
a calculation unit that calculates a linear combination of the read output signals and an optimal combination strength,
the signal conversion unit includes an optical amplifier that non-linearly converts the input signal, which is an optical signal, into the output signal, which is an optical signal;
The nonlinear conversion from the input signal to the output signal has reproducibility that if the input signal has the same waveform, the output signal also has the same waveform,
wherein the optical amplifier and the optical phase modulator are incorporated in a loop composed of an optical fiber;
a clock extractor that outputs an electrical signal indicating a clock frequency extracted from the optical signal output from the loop;
a phase shifter that inputs the electrical signal output from the clock extractor to the optical phase modulator;
has
The optical phase modulator corrects a phase change occurring in the loop based on the input electrical signal.
An information processing system characterized by: The signal conversion unit, the signal input unit, and the signal output unit form a loop,
2. The information processing system according to claim 1, wherein at least part of said output signal influences a signal that is subsequently input to said signal converter. An information processing method executed by an information processing system,
The information processing system is
a signal conversion unit that holds at least part of a signal input in the past as an internal state and non-linearly converts the input signal in the internal state;
a signal input unit that inputs the input signal to the signal conversion unit;
a signal output unit that reads an output signal that is a signal converted by the signal conversion unit;
a calculation unit that calculates a linear combination of the read output signals and an optimal combination strength,
the signal conversion unit includes an optical amplifier;
wherein the optical amplifier and the optical phase modulator are incorporated in a loop composed of an optical fiber;
a clock extractor that outputs an electrical signal indicating a clock frequency extracted from the optical signal output from the loop;
a phase shifter that inputs the electrical signal output from the clock extractor to the optical phase modulator;
has
wherein the optical amplifier non-linearly converts the input signal, which is an optical signal, to the output signal, which is an optical signal;
The nonlinear conversion from the input signal to the output signal has reproducibility that if the input signal has the same waveform, the output signal has the same waveform,
The optical phase modulator corrects the phase change occurring in the loop based on the input electrical signal.
An information processing method in an information processing system characterized by:

Images