Information processing device, information processing method, and information processing program

JPWO2024224540A5Pending Publication Date: 2025-09-29

View PDF 0 Cites 0 Cited by

Patent Information

Application Number: JP2025516399
Authority / Receiving Office: JP · JP
Patent Type: Applications
Filing Date: 2025-07-17
Publication Date: 2025-09-29

AI Technical Summary

Technical Problem

Current physical reservoir computing systems are limited by fixed weighting coefficients in the physical reservoir layer, preventing the switching of inference targets.

Method used

An information processing device and method that dynamically changes the internal connection state of the physical reservoir layer by applying actions, such as mechanical, optical, or magnetic effects, and adjusts the weighting coefficients of the output layer based on output values to enable switching of inference targets.

Benefits of technology

This approach allows for flexible switching of inference targets by modifying the internal coupling state of the physical reservoir layer, enhancing the system's ability to process different learning and inference tasks effectively.

✦ Generated by Eureka AI based on patent content.

Patent Text Reader

Abstract

An information processing device (1) includes: a neural network that includes an input layer (10), a physical reservoir layer (20) that serves as an intermediate layer, and an output layer (30); a function-imparting unit (40) that imparts a function to the physical reservoir layer (20) and thereby causes a change that corresponds to the function in the internal coupling state of the physical reservoir layer; an adjustment unit (31) that adjusts a weighting coefficient (Wj) for the output layer (30) on the basis of an output value from the output layer; and a control unit (50, 70) that controls the operation of the function-imparting unit (40).

Need to check novelty before this filing date? Find Prior Art

Description

Information processing device, information processing method, and information processing program

[0001] The present disclosure relates to an information processing device, an information processing method, and an information processing program.

[0002] A computing technology called reservoir computing, which is a type of hierarchical or interconnected neural network, has been proposed. In reservoir computing, the recurrently connected neural network is replaced with a reservoir layer, the weight coefficients of the reservoir layer are fixed, and learning is performed only in the output layer, thereby reducing the computational load.

[0003] Furthermore, a computing technology called physical reservoir computing has been proposed in which the reservoir layer is constructed from a physical reservoir layer (e.g., a physical material, an optical waveguide, etc.), and advanced processing such as time series prediction or data classification for the input of the input layer is realized by a small number of learning calculations in the output layer (see, for example, Patent Documents 1 to 3).

[0004] Japanese Patent No. 6701247 Japanese Patent No. 7108987 Japanese Patent Laid-Open No. 2022-80891

[0005] However, in current physical reservoir computing, the weighting coefficients of the physical reservoir layer are fixed, which means that it is not possible to switch the inference target.

[0006] The present disclosure aims to provide an information processing device, an information processing method, and an information processing program that are capable of switching an inference target.

[0007] The information processing device of the present disclosure is characterized by having a neural network including an input layer, a physical reservoir layer as an intermediate layer, and an output layer; an action imparting unit that imparts an action to the physical reservoir layer, thereby causing a change in the internal connection state of the physical reservoir layer corresponding to the action; an adjustment unit that adjusts the weight coefficient of the output layer based on the output value of the output layer; and a control unit that controls the operation of the action imparting unit.

[0008] The information processing method disclosed herein is a method executed by an information processing device having a neural network including an input layer, a physical reservoir layer as an intermediate layer, and an output layer, and an action imparting unit that imparts an action to the physical reservoir layer, and is characterized by having the steps of: adjusting the weight coefficient of the output layer based on the output value of the output layer; and causing the action imparting unit to impart an action to the physical reservoir layer, thereby causing a change in the internal coupling state of the physical reservoir layer corresponding to the action.

[0009] According to the present disclosure, the inference target can be switched by changing the internal bonding state of the physical reservoir layer.

[0010] 1 is a functional block diagram showing a configuration of an information processing device (learning device) according to a first embodiment. FIG. 2 is a diagram showing an example of a hardware configuration of the information processing device (learning device) according to the first embodiment. FIG. 3 is a schematic diagram showing how the state of a neural network in a physical reservoir layer changes due to an action. FIG. 4 is a schematic diagram showing how the state of a neural network in a physical reservoir layer changes due to a mechanical action (external force). (A) and (B) are schematic diagrams showing how the state of a neural network in a physical reservoir layer changes due to an optical action (light pattern). (A) and (B) are schematic diagrams showing how the state of a neural network in a physical reservoir layer changes due to a magnetic action (magnetic flux). FIG. 4 is a flowchart showing learning processing of the information processing device (learning device) according to the first embodiment. FIG. 5 is a functional block diagram showing a configuration of an information processing device (inference device) according to the first embodiment. FIG. 6 is a diagram showing an example of a hardware configuration of the information processing device (inference device) according to the first embodiment. FIG. 7 is a flowchart showing inference processing of the information processing device (inference device) according to the first embodiment. FIG. 8 is a functional block diagram showing another configuration of the information processing device (inference device) according to the first embodiment. FIG. 9 is a functional block diagram showing a configuration of an information processing device (learning device) according to a second embodiment. Fig. 1 is a flowchart showing a learning process of an information processing device (learning device) according to embodiment 2. Fig. 2 is a functional block diagram showing a configuration of an information processing device (inference device) according to embodiment 2. Fig. 3 is a flowchart showing an inference process of an information processing device (inference device) according to embodiment 2. Fig. 4 is a functional block diagram showing another configuration of an information processing device (inference device) according to embodiment 2.

[0011] An information processing device, an information processing method, and an information processing program according to embodiments will be described below with reference to the drawings. The following embodiments are merely examples, and the embodiments can be appropriately combined and modified. In addition, in the drawings, the same reference numerals are used to designate the same components or components having similar functions.

[0012] First Embodiment The first embodiment relates to an information processing device that enables switching of neural networks for different learning targets by changing the internal connection state of a certain physical reservoir layer.

[0013] 1 is a functional block diagram showing the configuration of an information processing device 1 (learning device) according to embodiment 1. The information processing device 1 is a device capable of implementing the information processing method (learning method) according to embodiment 1. A control unit 50 (learning control unit) of the information processing device 1 is, for example, a computer.

[0014] The information processing device 1 includes a neural network including an input layer 10, a physical reservoir layer 20 as an intermediate layer, and an output layer 30; an effect imparting unit 40 that imparts an effect to the physical reservoir layer 20, thereby causing a change in the internal connection state of the physical reservoir layer 20 corresponding to the imparted effect; an adjustment unit 31 that adjusts weight coefficients Wj (j is a positive integer assigned to multiple weight coefficients) of the output layer 30 based on the output value of the output layer 30; and a control unit 50 that controls the operation of the effect imparting unit 40. The information processing device 1 is connected to a storage unit 51 that is a storage device in which information is stored. The storage unit 51 may be a part of the information processing device 1.

[0015] In the first embodiment, in each of the neural networks switched by changing the action (i.e., in each of the multiple actions), the weight coefficient Wj of the output layer 30 is updated so as to obtain the desired inference result (e.g., the output value closest to the target value).

[0016] The control unit 50 uses learning data including learning input data input to the input layer 10 and target values that are learning teacher data to generate a learning model for inferring an output value that is an inference result from input information of an inference target that is input to the input layer 10. The control unit 50 stores the weight coefficient Wj for each of the multiple effects in the memory unit 51 as a parameter of the learning model.

[0017] Physical reservoir computing using a physical reservoir layer is described in, for example, Non-Patent Documents 1 and 2 listed below.

[0018] Kohei Tanaka, "Reservoir Computing (Keywords to Know, No. 136)," Journal of the Institute of Image Information and Television Engineers, Vol. 74, No. 3, pp. 532-534, 2020, https: / / www.ite.or.jp / contents / keywords / 2005keyword.pdf Satoshi Yamane, "Machine Learning Devices Using Physical Reservoir Computing and Their Applications to Edge Computing," ProVISON No. 95, pp. 61-65 (Technical Commentary 5), 2019, https: / / www.ibm.com / downloads / cas / 8MZNZZGX

[0019] The physical reservoir layer 20 converts information from the input layer 10 into a higher dimension using the inherent nonlinear properties of its material and structure, and is represented by a recursive neural network for time-series data. The neural network depends on the internal connection state, which is the connection state between virtual nodes in the physical reservoir layer 20.

[0020] In the first embodiment, the internal coupling state, which is the coupling state between the materials that make up the physical reservoir layer 20 and the virtual nodes within the structure thereof, is not kept fixed, but the internal coupling state is actively changed (including updated). By applying an action to the entire or part of the physical reservoir layer 20 using the action applying unit 40, it is possible to change the nonlinear characteristics (i.e., nonlinear dynamics) of the materials that make up the physical reservoir layer 20 and the structure of the physical reservoir layer 20 due to deformation or state changes. In other words, by changing the action applied to the physical reservoir layer 20, it is possible to switch the processing task of the information processing device 1 that includes the physical reservoir layer 20.

[0021] 2 is a diagram illustrating an example of a hardware configuration of an information processing device 1 (learning device) according to embodiment 1. The information processing device 1 according to embodiment 1 includes a processor 101 such as a CPU (Central Processing Unit), a memory 102 as a storage device such as a RAM (Random Access Memory), a storage device 103 as a volatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and an interface 104. These components correspond to the control unit 50 and storage unit 51 in FIG. 1. These components may also be configured using dedicated processing circuits.

[0022] The processor 101 can execute the information processing program according to the first embodiment. The information processing program is recorded on a recording medium such as an SD (Secure Digital) memory card or a USB (Universal Serial Bus) memory card, or is acquired by downloading via a network. Note that the hardware configuration shown in FIG. 2 is an example, and various modifications to the hardware configuration are possible.

[0023] 3 is a schematic diagram showing how the state of the neural network in the physical reservoir layer 20 changes due to an action. The internal bonding state of the physical reservoir layer 20 can be changed by applying an action by the action applying unit 40 to the entire or part of the acted surface of the physical reservoir layer 20. It is also possible to employ a physical reservoir layer 20 in which the internal bonding state of the physical reservoir layer 20 when the action by the action applying unit 40 is applied to the entire acted surface of the physical reservoir layer 20 is different from the internal bonding state of the physical reservoir layer 20 when the action by the action applying unit 40 is applied to a part of the acted surface of the physical reservoir layer 20 (e.g., half a region, a quarter a region, a circular region, a square region, a triangular region, etc.). Furthermore, in cases where no action is required on a portion of the acted surface of the physical reservoir layer 20, it is also possible to employ a physical reservoir layer 20 in which the internal bonding state differs depending on, for example, the shape or area of the area to which the action of the acted surface is applied (e.g., half area, 1 / 4 area, circular area, square area, triangular area, etc.). The action includes at least one element of light, electricity, magnetism, heat, external force, and chemical substance applied to the physical reservoir layer 20.

[0024] FIG. 4 is a schematic diagram showing that the state of the neural network in the physical reservoir layer 20 changes due to a mechanical action (external force).

[0025] 5A and 5B are schematic diagrams showing how the state of a neural network formed by the physical reservoir layer 20 changes due to an optical effect (light pattern). In the examples of FIGS. 5A and 5B, the effect imparting unit 40 is a light irradiation device that irradiates the physical reservoir layer 20 with light. In FIG. 5A, patterned light is irradiated onto a portion of the physical reservoir layer 20, and the internal bonding state can be made different depending on the region of the physical reservoir layer 20 to which the effect (light pattern) is imparted. In other words, in FIG. 5A, the internal bonding state of the region irradiated with the patterned light is different from the internal bonding state of the region not irradiated with the patterned light. The patterned light can be generated by an optical element having optical filter characteristics that transmit, reflect, or absorb light from a light source in a portion of its region. Alternatively, the patterned light can be generated by using a spatial light modulation element that can dynamically switch the light transmission, reflection, or absorption characteristics. The physical reservoir layer 20 may also convert input light into intermediate light of speckle light by repeatedly scattering and diffusing light therein. The input light to the physical reservoir layer 20 may also be speckle light. This is because the interference pattern of speckle light has the characteristic of changing sensitively depending on the interference state.

[0026] Furthermore, the physical system of the physical reservoir layer 20 may contain molecules or particles as contained substances therein, and may constitute a neural network optically coupled by the contained substances.

[0027] 6A and 6B are schematic diagrams showing how the state of the neural network in the physical reservoir layer 20 changes due to magnetic action (magnetic flux). In the examples of FIGS. 6A and 6B, the action imparting unit 40 is a coil that serves as a magnetic flux imparting unit that generates magnetic flux (e.g., magnetic flux density distribution pattern 41) imparted to the physical reservoir layer 20. Similar to the optical action shown in FIG. 5A, FIG. 6B shows an example in which the state of the neural network formed by the physical reservoir layer 20 is changed by a magnetic flux density distribution pattern that imparts bias to the magnetic flux density distribution or regions of high or low magnetic flux density (i.e., patterned magnetic flux density). In other words, the internal coupling state in regions with strong magnetic flux differs from the internal coupling state in regions with weak magnetic flux. The number of coils used to impart a magnetic flux density distribution pattern may be three or more.

[0028] Below, examples of actions on the physical reservoir layer 20 of the information processing device 1 in embodiment 1 are shown. Mechanical actions include pressure, deformation, torsional deformation, etc. Optical actions include light intensity, wavelength, optical plasma, etc. Electrical actions include electric field, current, voltage, charge, etc. Thermal actions include heat absorption, heating, cooling, etc. Magnetic actions include magnetic field, magnetic force, etc. Chemical actions include liquid molecules, gas molecules, etc.

[0029] 7 is a flowchart showing the learning process of the information processing device 1 (learning device) according to embodiment 1. In the learning process of the information processing device 1, an action is applied to the physical reservoir layer 20 and learning data is input to the input layer 10 (step S11), weight coefficients Wj are learned as parameters of the learning model (step S12), and the weight coefficients as parameters are output (step S13). The output weight coefficients Wj are stored in the storage unit 51.

[0030] Furthermore, a learning model may be used to amplify the difference between two difficult-to-distinguish patterns and convert them into a pattern that is easier to distinguish. In this case, learning is performed while changing the weighting coefficients to determine the weighting coefficients as parameters of the learning model, and then a first action is obtained by applying an action to the physical reservoir layer 20 to infer a first target value for a first inference object, and a second action is obtained by applying another action to the physical reservoir layer 20 to obtain a second target value for a second inference object, and a learning model can be generated that can determine whether input patterns are identical based on the first and second actions.

[0031] FIG. 8 is a functional block diagram showing the configuration of an information processing device 1a (inference device) according to the first embodiment. The information processing device 1a is a device capable of implementing the information processing method (inference method) according to the first embodiment. The control unit 60 (inference control unit) of the information processing device 1a is, for example, a computer. The control unit 60 of the information processing device 1a infers an output value of an inference result output from the output layer 30 from input information of an inference target input to the input layer 10 using a learning model corresponding to the action assigned by the action assigning unit 40. Note that the memory unit 61 stores the learning model of the memory unit 51 in FIG. 1, but the memory unit 61 may be the same as the memory unit 51.

[0032] Furthermore, the control unit 60 may select a learning model corresponding to the effect imparted by the effect imparting unit 40 from learning models for inferring an output value from input information input to the input layer 10, which are generated by a learning device that uses learning data including learning input data and learning teacher data, and may use the selected learning model to infer the output value of the inference result output from the output layer 30 from the input information of the inference target input to the input layer 10. In this case, the information processing device 1a is a learning / inference device.

[0033] 9 is a diagram showing an example of the hardware configuration of an information processing device 1a (inference device) according to embodiment 1. The information processing device 1a according to embodiment 1 has a processor 111 such as a CPU, a memory 112 as a storage device such as RAM, a storage device 113 as a volatile storage device such as an HDD or SSD, and an interface 114. These components correspond to the control unit 60 and storage unit 61 in FIG. 8. Furthermore, these components may be configured using dedicated processing circuits.

[0034] The processor 111 can execute the information processing program according to the first embodiment. The information processing program is recorded on a recording medium such as an SD memory card or a USB memory card, or is acquired by downloading via a network. Note that the hardware configuration shown in Fig. 9 is an example, and various modifications to the hardware configuration are possible.

[0035] 10 is a flowchart showing the inference process of the information processing device 1a (inference device) according to embodiment 1. In the inference process of the information processing device 1, an action is applied to the physical reservoir layer 20, and input data to be inferred is input to the input layer 10 (step S21), and inference is performed using parameters (weighting coefficients) of the learning model (step S22), and the inference result is output (step S23).

[0036] Furthermore, when identifying the difference between two difficult-to-distinguish patterns, the control unit 60 may first switch the inference to a first inference object, load a condition for reproducing a first action state, assign the first action state to the physical reservoir layer, and execute the inference. Next, the control unit 60 may switch the inference to a second inference object, load a condition for reproducing a second action state, assign the second action state to the physical reservoir layer, execute the inference, and make a judgment based on the two inference results.

[0037] 11 is a functional block diagram showing another configuration of an information processing device 1b (inference device) according to embodiment 1. The information processing device 1b is a device capable of implementing the information processing method (inference method) according to embodiment 1. The control unit 60 (inference device) of the information processing device 1b is, for example, a computer. The control unit 60 can perform inference based on a learning model using the input information and output values without updating the weight coefficients Wj of the output layer 30 so as to obtain an output value closest to a target value.

[0038] As described above, the information processing devices 1, 1a, and 1b based on physical reservoir computing can accommodate a variety of learning and inference processes by changing their functions. Furthermore, the information processing devices 1, 1a, and 1b enable highly sensitive sensing based on physical reservoir computing.

[0039] Second Embodiment The second embodiment is an information processing device that switches neural networks for different learning targets by changing the internal connection state of a certain physical reservoir layer 20, and makes it possible to estimate feature quantities that change the internal connection state of the physical reservoir layer 20 from changes in the weight coefficients of the output layer updated so that a desired inference result is obtained via the neural network. In the second embodiment, a usage method that differs from the conventional concept of physical reservoir computing is adopted as follows.

[0040] Unlike the first embodiment, in the second embodiment, the internal connection state, which is the connection state of the virtual nodes of the neural network, changes when an action is applied to the physical reservoir layer 20, and the mode of the action is estimated from the change in the weight coefficient Wj of the output layer 30 after learning that occurs as a result of this change. In other words, in the second embodiment, the action applied to the physical reservoir layer 20 is input information as an inference target, and the output data based on the weight coefficient Wj is the inference result.

[0041] Furthermore, in the second embodiment, there are generally multiple weighting coefficients Wj, which is advantageous when distinguishing between the characteristics of the first state and the characteristics of the second state based on the information on the weighting coefficients Wj, which are part of the inference process, compared to when distinguishing between the characteristics of the first state and the characteristics of the second state based on the final output result (inference result) of the output layer 30.

[0042] In addition, in the second embodiment, in learning the weight coefficient Wj of the output layer 30, which is performed to find the change in the action characteristics between the first state and the second state, a certain reference input data group and a corresponding teacher data group may be used as the input information as learning data.

[0043] Furthermore, the teacher data group may be switched to one of a plurality of different teacher data groups. In this case, each different teacher data group can be used to switch the range or type of action characteristics to be sought.

[0044] 12 is a functional block diagram showing the configuration of an information processing device 2 (learning device) according to embodiment 2. The information processing device 2 is a device capable of implementing the information processing method (learning method) according to embodiment 2. A control unit 70 (learning control unit) of the information processing device 2 is, for example, a computer.

[0045] The information processing device 2 has a neural network including an input layer 10, a physical reservoir layer 20 as an intermediate layer, and an output layer 30, an effect imparting unit 40 that imparts an effect to the physical reservoir layer 20 to cause a change in the internal connection state of the physical reservoir layer 20 corresponding to the imparted effect, an adjustment unit 31 that adjusts weight coefficients Wj (j is a positive integer assigned to multiple weight coefficients) of the output layer 30 based on the output value of the output layer 30, and a control unit 70 that controls the operation of the effect imparting unit 40. The information processing device 2 is connected to a storage unit 71 that is a storage device in which information is stored. The storage unit 71 may be a part of the information processing device 2.

[0046] The control unit 70 of the information processing device 2 uses learning data including learning action data and learning teacher data to generate a learning model for inferring an inference result based on the weight coefficient Wj from the action imparted to the physical reservoir layer 20, and stores the weight coefficient Wj for each of the multiple actions as a parameter of the learning model in the memory unit 71. The hardware configuration of the information processing device 2 is the same as that of the first embodiment.

[0047] 13 is a flowchart showing the learning process of the information processing device 2 (learning device) according to embodiment 2. In the learning process of the information processing device 2, learning input data is input to the input layer 10 and acts as learning data in the physical reservoir layer 20 (step S31), weight coefficients Wj are learned as parameters of the learning model (step S32), and the weight coefficients as parameters are output (step S33). The output weight coefficients Wj are stored in the storage unit 71.

[0048] Furthermore, a learning model may be used to amplify the difference between two difficult-to-distinguish patterns and convert them into patterns that are easier to distinguish. In this case, a learning model can be generated by applying an action to the physical reservoir layer 20 to set the internal connection state of the physical reservoir layer 20 to a first state, changing the weight coefficients, executing learning, obtaining a first weight coefficient set Wg1, applying another action to the physical reservoir layer 20 to set the internal connection state of the physical reservoir layer 20 to a second state, changing the weight coefficients, executing learning, obtaining a second weight coefficient set Wg2, and comparing the first weight coefficient set Wg1 and the second weight coefficient set Wg2 to determine whether the input actions are the same.

[0049] FIG. 14 is a functional block diagram showing the configuration of an information processing device 2a (inference device) according to the second embodiment. The information processing device 2a is a device capable of implementing the information processing method (inference method) according to the second embodiment. The control unit 60 (inference control unit) of the information processing device 2a is, for example, a computer. The control unit 80 of the information processing device 2a uses a learning model corresponding to the effect imparted by the effect imparting unit 40 to infer an output value of an inference result consisting of a weighting coefficient from the effect of the inference target. The data processing unit 82 outputs the inference result from the weighting coefficient. The inference result is stored in the memory unit 81. Note that the memory unit 81 stores the learning model of the memory unit 71 in FIG. 12, but the memory unit 81 may be the same as the memory unit 71. The data processing unit 82 can calculate an effect amount indicating the strength of the effect imparted to the physical reservoir layer based on the difference between a first weighting coefficient, which is a weighting coefficient when the internal bond state of the physical reservoir layer 20 is in a first state, and a second weighting coefficient, which is a weighting coefficient when the internal bond state of the physical reservoir layer is in a second state different from the first state.

[0050] Furthermore, the control unit 60 may select a learning model corresponding to the effect imparted by the effect imparting unit 40 from learning models for inferring an output value from input information input to the input layer 10, which are generated by a learning device that uses learning data including learning input data and learning teacher data, and may use the selected learning model to infer the output value of the inference result output from the output layer 30 from the input information of the inference target input to the input layer 10. In this case, the information processing device 1a is a learning / inference device.

[0051] 15 is a flowchart showing the inference process of the information processing device 2 a (inference device) according to embodiment 2. In the inference process of the information processing device 2 a, input data is input to the input layer 10, and the action of the inference target is assigned to the physical reservoir layer 20 (step S41), inference is performed using the parameters (weighting coefficients) of the learning model (step S42), and the inference result is output (step S43).

[0052] 16 is a functional block diagram showing another configuration of an information processing device 2b (inference device) according to embodiment 2. In this case, the control unit 80 updates the learning model based on the effect imparted by the effect imparting unit 40 and the inference result, without updating the weight coefficients Wj of the output layer 30 so as to obtain an output value closest to the target value.

[0053] As described above, the information processing devices 2, 2a, and 2b based on physical reservoir computing can handle a variety of learning and inference processes by targeting actions as the inference target. Furthermore, the information processing devices 2, 2a, and 2b enable highly sensitive sensing based on physical reservoir computing.

[0054] The information processing devices of the first and second embodiments and their variations (hereinafter simply referred to as "the above embodiments") may be modified as appropriate. For example, components of the physical reservoir layer 20 and the action adding unit 40 of the above embodiments may be changed, added, or deleted. Furthermore, the features or components of the above embodiments may be combined as appropriate in different ways than those described above.

[0055] 1, 2 Information processing device (learning device), 1a, 1b, 2a, 2b Information processing device (inference device), 10 Input layer, 20 Physical reservoir layer (intermediate layer), 30 Output layer, 40 Action imparting unit, 50, 70 Control unit (learning control unit), 51, 71 Memory unit, 60, 80 Control unit (inference control unit), 61, 81 Memory unit, Wj Weight coefficient.

Claims

1. A neural network including at least an input layer, a physical reservoir layer as an intermediate layer, and an output layer; an action imparting unit that imparts an action to the physical reservoir layer; an adjustment unit that adjusts weight coefficients of the output layer based on the output values of the output layer; A control unit that controls the operation of the action imparting unit; An information processing device comprising:

2. The action imparting unit imparts an action to the physical reservoir layer, thereby causing a change in the internal bonding state of the physical reservoir layer corresponding to the action.