Information processor

Abstract

To provide an information processor capable of improving accuracy of estimation of an animal behavior and neural activities by machine learning and being applied to a wide application range.SOLUTION: An information processor acquires estimation information which represents behaviours or neural activities of a predetermined target animal by using a machine learning model which has performed machine learning of a relationship between information about a biological signal acquired from the target animal and output information representing behaviors or neural activities of the target animal, and outputs the estimation information.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an information processing device using a machine learning model. [Background technology]

[0002] In machine learning models that predict animal behavior, sensory stimuli have traditionally been used as input and behavior-specific information has been used as output. However, conventional machine learning processing has not yet been able to predict behavior with sufficient accuracy. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] RCFong, et al., Using human brain activity to guide machine learning, Scientific Reports, 2018, DOI:10.1038 / s41598-018-23618-6 Summary of the Invention [Problem to be solved by the invention]

[0004] Meanwhile, in recent years, examples are being considered in which fMRI data representing human brain activity information is used for machine learning processing (Non-Patent Document 1).

[0005] The technology of Non-Patent Document 1 uses fMRI as brain activity information, which is used for machine learning with a support vector machine. For this purpose, the technology of Non-Patent Document 1 vectorizes data obtained by fMRI and uses it for preprocessing of machine learning with a support vector machine for classification processing. For these reasons, the technology of Non-Patent Document 1 has limited applications and cannot be used, for example, to estimate behavior or neural activity.

[0006] The present invention has been made in consideration of the above-mentioned situation, and one of its objectives is to provide an information processing device that can improve the accuracy of estimating animal behavior and neural activity using machine learning and can be applied to a wide range of applications. [Means for solving the problem]

[0007] One aspect of the present invention for solving the problems of the above-mentioned conventional examples is an information processing device that includes: input means for accepting input of biosignal information representing biosignals output by nerve cells of a target animal and information representing the behavior or neural activity of the target animal; estimation means for executing a predetermined estimation process using a machine learning model that has been machine-learned to understand the relationship between the biosignal information accepted by the input means and the information representing the behavior or neural activity of the target animal that has been input together with the input information; and output means for outputting the result of the estimation process by the estimation means. Effect of the Invention

[0008] According to the present invention, it is possible to improve the accuracy of estimating animal behavior and neural activity through machine learning, enabling a wide range of applications. [Brief description of the drawings]

[0009] [Figure 1] 1 is a configuration block diagram illustrating an example of an information processing device according to an embodiment of the present invention. [Diagram 2] 1 is a functional block diagram illustrating an example of an information processing device according to an embodiment of the present invention. [Diagram 3] 11 is an explanatory diagram illustrating an example of a result of an estimation process using the information processing device according to the embodiment of the present invention. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] An embodiment of the present invention will be described with reference to the drawings. One aspect of an information processing device 1 according to an embodiment of the present invention includes a control unit 11, a storage unit 12, an operation unit 13, a display unit 14, an input / output unit 15, and a biosignal sensor unit 16 connected via the input / output unit 15, as shown in FIG.

[0011] [Basic configuration] The control unit 11 is a program-controlled device such as a CPU, and operates according to a program stored in the storage unit 12. In this embodiment, the control unit 11 executes machine learning processing and estimation processing. The control unit 11 that performs the machine learning processing receives input of a biosignal of a target animal (which may be a human or a non-human animal; hereinafter, referred to as a target animal) from the biosignal sensor unit 16, A / D converts (digitizes) the biosignal to information related to the biosignal (hereinafter, referred to as biosignal information), and obtains input information including the biosignal information.

[0012] One of the features of this embodiment is that the biosignal information is information obtained by directly A / D converting the biosignal.

[0013] The control unit 11 inputs input information including the biological signal information to a predetermined neural network, obtains output information of the neural network, and performs machine learning on the neural network so that the output information approaches information (called teacher information) that represents the behavior or neural activity of the target animal. This machine learning may be performed by adopting a general machine learning method that controls weights and biases between elements in the neural network based on the difference between the output information and the teacher information.

[0014] When performing the estimation process, the control unit 11 receives an input of a biosignal from the biosignal sensor unit 16. The biosignal received here is the same type of biosignal as that input during the machine learning process. Here, the same type of biosignal means, for example, if the biosignal used during the machine learning process is an electrocardiogram signal, the same electrocardiogram signal is used, and if the biosignal used during the machine learning process is a signal output by a certain nerve cell of the target animal, the nerve cell or the same type of nerve cell as the nerve cell (nerve cells of the same tissue, or cells belonging to the same sensory or motor area in the case of the brain, etc.) (hereinafter the same). The control unit 11 A / D converts (digitizes) the received biosignal to biosignal information related to the biosignal, and obtains input information including the biosignal information.

[0015] The control unit 11 receives input information including biosignal information related to the same type of biosignal as the biosignal obtained by the machine learning process, etc., and uses a machine learning model (the neural network) that has learned the relationship between the input information and output information representing the behavior or neural activity of the target animal by machine learning, inputs the input information including biosignal information related to the biosignal to the machine learning model, and obtains estimated information representing the behavior or neural activity of the target animal as the output of the machine learning model.The control unit 11 then outputs the estimated information obtained as a result of the estimation.

[0016] Here, the type of signal received by the control unit 11 is not limited to that input from the biosignal sensor unit 16, and may further include stimulation information representing the content of the stimulation for the target animal. In this case, the biosignal received by the control unit 11 from the biosignal sensor unit 16 and used as the input signal is a biosignal acquired after the target animal receives the stimulation represented by the stimulation information. In this example, the control unit 11 inputs input information including at least the stimulation information and biosignal information related to the biosignal obtained by A / D converting (digitizing) the received biosignal during the machine learning process and the estimation process to the neural network, and executes the machine learning process and the estimation process. The operation of the control unit 11 will be described in detail later.

[0017] The storage unit 12 is a memory device or a disk device, and holds a program executed by the control unit 11. This program may be provided by being stored in a computer-readable and non-transitory recording medium, and may be copied to the storage unit 12. The storage unit 12 also operates as a work memory for the control unit 11.

[0018] In one example of this embodiment, the memory unit 12 holds information related to the neural network used by the control unit 11 in machine learning processing and estimation processing (such as information related to the neural network model and information such as weights and biases between elements that make up the neural network).

[0019] The operation unit 13 is a mouse, a keyboard, etc., which accepts operations from a user and outputs information representing the operations to the control unit 11. The display unit 14 is a display, etc., which displays and outputs information according to instructions input from the control unit 11 to present it to the user.

[0020] The input / output unit 15 includes a network interface, a serial interface, etc. The input / output unit 15 outputs a signal input from the biosignal sensor unit 16 to the control unit 11. A camera C that captures the behavior of the target animal may be connected to the input / output unit 15. Furthermore, the input / output unit 15 outputs information related to the instruction to an external computer device or the like in accordance with the instruction input from the control unit 11.

[0021] [An example of a machine learning model] Next, an example of a machine learning model used by the information processing device 1 of this embodiment will be described. The machine learning model used by the control unit 11 of this embodiment may be a general neural network having at least three layers, namely, an input layer, an intermediate layer, and an output layer.

[0022] Since the biosignal information relating to the biosignal input from the biosignal sensor unit 16 is generally time-series information (information that changes over time), it is preferable that this machine learning model be a neural network capable of at least machine learning the time-series information, such as a recurrent network or a reservoir network.

[0023] In the following example of this embodiment, the machine learning model used by the control unit 11 is described as a reservoir network (e.g., W. Maass et al., Real-time computing without stable states: a new framework for neural computation based on perturbations, Neural comput. Vol.14 No.11, pp.2531-2560, 2002 doi:10.1162 / 089976602760407955), but as mentioned above, this is just one example, and other types of neural networks may also be used.

[0024] [Example of biosignals] Here, the sensor S included in the biosignal sensor unit 16 detects the electrical signal of the nerve cells of the target animal as a biosignal. As an example, the sensor S may be an invasive method, that is, a method of directly detecting the electrical signal of a single nerve cell. In this case, the biosignal sensor unit 16 has a plurality of sensors S (generally 100 or more), and each sensor S directly detects an electrical signal output from one of the different nerve cells.

[0025] As another example, the sensor S may be one that detects signals based on electrical signals of nerve cells in a non-invasive manner. As the sensor S using such a non-invasive method, various widely known ones that detect signals of nerve cells in the human brain, such as electroencephalogram (EEG), magnetoencephalogram (MEG), single photon CT (SPECT), positron CT (PET), functional MRI (fMRI), near-infrared spectroscopy (NIRS), transcranial magnetic stimulation (TMS), etc., can be adopted.

[0026] In this embodiment, the above-mentioned nerve cells are not limited to brain cells, but may be spinal nerves or the like.

[0027] [Control unit operation] Next, the operation of this embodiment will be described. The control unit 11 executes a program stored in the storage unit 12 to realize a configuration that functionally includes at least one of a machine learning processing unit 30 and an estimation processing unit 40, as exemplified in FIG.

[0028] Here, the machine learning processing unit 30 functionally includes a signal receiving unit 31, a teacher information receiving unit 32, and a machine learning execution unit 33. The estimation processing unit 40 functionally includes a signal receiving unit 41, an estimation execution unit 42, and an output processing unit 43.

[0029] The signal receiving unit 31 sequentially (at a predetermined timing) receives input of detected biosignals from the biosignal sensor unit 16, which sequentially (repeatedly at a predetermined timing) detects and outputs biosignals using a sensor S attached to the target animal. The teacher information receiving unit 32 sequentially receives information representing the behavior or neural activity of the target animal as teacher information. Here, the information representing behavior may be, for example, information representing the movements (trajectories) of multiple body parts of the target animal. For example, if the target animal is a mouse, this may be the movement trajectory (two-dimensional movement) of the nose as seen from the front of the mouse, and if n body parts are targeted, this will be at least 2n-dimensional vector information.

[0030] However, the behavior of the target animal here does not necessarily have to be something that can be seen from the outside, and it may be an involuntary movement rather than a voluntary movement. Specifically, it may be the heart rate, the number of breaths, etc.

[0031] In one example of the present embodiment, when the information representing these actions is multi-dimensional vector information, the dimensions may be reduced by performing principal component analysis.

[0032] Furthermore, the information representing the neural activity may be, for example, image information or audio information representing an image captured by the target animal in its visual field or a sound captured by its hearing. In this case, the information representing the neural activity does not necessarily need to be information obtained via the optic nerve or auditory nerve of the target animal, but may be image information obtained by a camera capturing an image in the same direction as the line of sight of the target animal, or audio information collected by a microphone attached near the ear of the target animal (that is, it may be information itself representing a stimulus given to the target animal).

[0033] The machine learning execution unit 33 executes machine learning processing of the machine learning model stored in the storage unit 12. Specifically, the machine learning execution unit 33 sequentially A / D converts (digitizes) the biosignals sequentially received by the signal receiving unit 31 to generate sequential biosignal information related to the biosignals. In this embodiment, the machine learning execution unit 33 acquires information representing a stimulus given to the target animal while the signal receiving unit 31 is receiving the biosignals.

[0034] The information representing the stimulus may be audio information made to the target animal, or, if the target animal is a mouse or the like, may represent the operation of the target animal on an operation object such as a spout lever from which a reward can be obtained in response to a specified operation, or a reward given to the target animal.

[0035] The machine learning execution unit 33 sequentially inputs the sequentially obtained biological signal information and information representing the stimulation to the machine learning model as input information of the machine learning model that is the target of machine learning.

[0036] The machine learning execution unit 33 sequentially obtains the output of the machine learning model, compares the obtained output of the machine learning model (here, it is assumed that the output is of the same dimension as the information representing the behavior or neural activity of the target animal, which is the teacher information) with the teacher information sequentially input, and updates the weights and biases of the machine learning model based on the difference so as to reduce the difference (general backpropagation processing may be used). Here, when the machine learning model is a reservoir network, the machine learning execution unit 33 updates the connection weights between the intermediate layer, which is the reservoir layer, and the output layer.

[0037] In this embodiment, input information including information (biological signal information) relating to a signal that changes in a time series of the same type as the signal that is sequentially output by the biosignal sensor unit 16 is sequentially input, and a machine learning model is obtained in which the relationship between the input information and output information representing the behavior or neural activity of the target animal has been machine learned.

[0038] Next, the operation of the estimation processing unit 40 of the control unit 11 will be described. The signal receiving unit 41 of the estimation processing unit 40 sequentially receives input of detected biological signals from the biological signal sensor unit 16, which sequentially detects and outputs biological signals by a sensor S attached to a target animal. Here, the target animal may be the same individual as the individual from which the biological signal was obtained when the processing of the machine learning processing unit 30 was executed.

[0039] The estimation execution unit 42 sequentially A / D converts (digitizes) the biosignals sequentially received by the signal receiving unit 41 to generate sequential biosignal information related to the biosignals. The estimation execution unit 42 also acquires information representing a stimulus given to the target animal while the signal receiving unit 41 is receiving the biosignals. Here, the information representing the stimulus uses the same type of information as when the machine learning model is trained by the operation of the machine learning processing unit 30.

[0040] The estimation execution unit 42 sequentially inputs the above-mentioned sequentially obtained biosignal information and information representing the stimulus to the machine learning model that has been machine-learned by the operation of the above-mentioned machine learning processing unit 30 as input information for the machine learning model that is the subject of machine learning.

[0041] The estimation execution unit 42 then sequentially obtains the output of the machine learning model, and outputs the obtained output of the machine learning model to the output processing unit 43.

[0042] The output processing unit 43 outputs the output of the machine learning model sequentially output by the estimation execution unit 42 as estimated information representing the behavior or neural activity of the target animal, by displaying the information related to the behavior or neural activity of the target animal represented by the content on the display unit 14, etc.

[0043] [Operation] The information processing device 1 of this embodiment basically has the above configuration and operates as follows. As an operation example, an example of estimating the behavior of a mouse (movement of each part of the mouse) when audio information and a stimulus from a spout lever are given to the mouse will be described.

[0044] The biosignal sensor unit 16 of the information processing device 1 in this example has a plurality of (e.g., about 300) sensors S, each of which is invasively connected to different brain nerve cells of the mouse. The information processing device 1 then executes machine learning processing as follows, using biosignals (here, electrical signals generated by each brain nerve cell to which the sensor S is connected) sequentially detected by the sensors S of the biosignal sensor unit 16.

[0045] The information processing device 1 obtains biosignals successively detected by the sensor S of the biosignal sensor unit 16 while providing different audio signal stimuli at different times to the mouse connected to the sensor S (and providing intermittent stimuli by the spout lever according to the mouse's voluntary actions). The information processing device 1 also sequentially obtains information representing the mouse's behavior as teacher information based on images captured by a camera C or the like connected to the input / output unit 15. In this example, the movement trajectory (two-dimensional movement) of the mouse's nose as viewed from the front is used, and 56-dimensional vector information is obtained for 28 body parts.

[0046] Furthermore, the information processing device 1 inputs, as input information, time-series information representing the applied stimulation and biosignal information obtained by digitizing the sequentially detected biosignals to a machine learning model that is a reservoir network.

[0047] The information processing device 1 sequentially obtains the output of the machine learning model for the above input information, and uses the obtained output of the machine learning model and the sequentially obtained teacher information to update the weights of the machine learning model so as to reduce the difference between them (backpropagation processing).

[0048] By executing the above process for a certain period of time, the information processing device 1 sequentially inputs input information including information (biological signal information) relating to a signal that changes in a time series of the same type as the signal sequentially output by the biological signal sensor unit 16, and obtains a machine learning model in which the relationship between the input information and output information representing the behavior of the target animal (the individual that is the subject of machine learning) has been machine learned.

[0049] That is, in this example of the present embodiment, the information processing device 1 obtains a machine learning model that mimics the response of the brain nerves of an individual mouse in which electrical signals from brain nerve cells have been detected in the machine learning processing.

[0050] Then, the information processing device 1 uses this machine learning model to estimate the behavior of the mouse (the individual mouse from which the electrical signals of the brain nerve cells have been detected in the machine learning process) in response to the stimulus as follows.

[0051] That is, the information processing device 1 acquires the bio-signals detected sequentially by the sensor S of the bio-signal sensor unit 16 while providing different audio signal stimuli at different times to the mouse connected to the sensor S (and also providing intermittent stimuli via the spout lever in response to the mouse's voluntary actions).

[0052] The information processing device 1 then inputs, as input information, time-series information representing the applied stimulation and biosignal information obtained by digitizing the sequentially detected biosignals to a machine learning model that is a reservoir network.

[0053] The information processing device 1 sequentially obtains the output of the machine learning model for the above input information, and outputs the obtained output of the machine learning model as information representing the behavior of the mouse.

[0054] The information processing device 1 in this example of the present embodiment performs machine learning processing on, for example, an individual mouse that has no impairment in the movement of each part of the body, and obtains a machine learning model, which can be used to estimate the movement that the individual mouse intended to perform when a movement impairment occurs later.

[0055] An experimental example showing to what extent the information processing device 1 of this embodiment is capable of estimating the behavior of an individual mouse will be described later.

[0056] [Application example 1. Disease model] In the information processing device 1 of this embodiment, as described above, a signal based on the activity of an individual's brain neurons is detected by the sensor S of the biosignal sensor unit 16 (whether invasive or non-invasive) and used as input for a machine learning model, making it possible to estimate the response of the individual's brain neurons to some kind of stimulus.

[0057] Therefore, if the information processing device 1 obtains the above machine learning model by machine learning when no disability occurs, if the individual that was the subject of the machine learning later suffers from brain damage, the information processing device 1 can temporarily set some of the weights between the reservoir layer and the output layer of the reservoir network in the machine learning model to ``0'', execute an estimation process, and compare the output with the actual behavior of the individual.It is expected that understanding of the individual's illness and knowledge of healing actions can be provided depending on which weight was set to ``0'' when the behavior of the individual could be best estimated.

[0058] Similarly, instead of setting the weight to "0", the information processing device 1 may execute a process of delaying the propagation of at least a part of information between the input layer and the reservoir layer of the machine learning model, or at least a part of information within the reservoir layer, or at least a part of information between the reservoir layer and the output layer, compared to other parts. This is expected to provide knowledge of the individual's illness and healing behavior if it is possible to know which part should be delayed to best estimate the individual's behavior.

[0059] [Application Example 2. Visualization of hallucinations and auditory hallucinations] Furthermore, in this embodiment, the information processing device 1 may detect and digitize a signal based on the activity of brain nerve cells of an individual (whether invasive or non-invasive) using the sensor S of the biosignal sensor unit 16, use the biosignal information as input information for a machine learning model, and train the above machine learning model using visual stimuli (image information) given to the individual while the biosignal is being detected as teaching information.

[0060] The machine learning model thus machine-learned can be used as follows: That is, the information processing device 1 detects and digitizes a signal based on the activity of the brain nerve cells of the individual with the sensor S of the biosignal sensor unit 16, and inputs the biosignal information as input information to the machine learning model. The machine learning model estimates and outputs image information that the individual is seeing (not whether or not it is actually seen, but what the brain recognizes) from the signal based on the activity of the brain nerve cells.

[0061] According to this example, for an individual experiencing hallucinations, for example, it becomes possible to estimate and reproduce the hallucinations that the individual is experiencing. Also, if audio information is used instead of visual stimuli (image information), it becomes possible to estimate and reproduce auditory hallucinations.

[0062] [Application example 3. Assistance for machine operation] In the explanation so far, it has been explained that the target animal (individual) from which the biosignal is obtained during the machine learning processing and the target animal from which the biosignal is obtained during the estimation processing are the same individual, but this embodiment is not limited to this (they may be the same individual, but this is not necessary).

[0063] For example, the information processing device 1 regards visual and auditory stimuli (image information and audio information) given to a human (individual) performing some kind of machine operation (e.g., operating a game controller or driving a car, etc.) as information representing stimuli. The information processing device 1 also detects and digitizes signals based on the activity of the individual's brain nerve cells (whether invasive or non-invasive) using the sensor S of the biosignal sensor unit 16 to obtain the biosignal information. The information processing device 1 then uses the information representing these stimuli and the biosignal information based on the biosignals as input signals, and uses information representing the content of the machine operation of the individual while the biosignals are being obtained (the stimuli are being given) as teacher information to machine-train the machine learning model.

[0064] The machine learning model that has been machine-learned according to this example can be used as follows: That is, the information processing device 1 acquires visual and auditory stimuli (image information and audio information) given to the individual or a human individual different from the individual (referred to as an estimated subject for distinction) as information representing the stimuli, and detects and digitizes signals based on activity of the brain nerve cells of the estimated subject with the sensor S of the biosignal sensor unit 16 to obtain the biosignal information.

[0065] The information processing device 1 inputs the information representing these stimuli and the biosignal information based on the biosignal as input signals to a machine learning model, obtains an output of the machine learning model, and provides assistance with machine operation. Here, the assistance may be, for example, an operation signal of a controller output without manual operation by the estimated subject, or, in the case of driving operation, information that prompts (guidances) the estimated subject to perform the estimated driving operation may be provided.

[0066] [Application Example 4. Estimation of the effects of drugs] In yet another example, the information processing device 1 of this embodiment detects and digitizes a signal based on the activity of brain nerve cells of an individual (regardless of whether it is invasive or non-invasive) using the sensor S of the biosignal sensor unit 16 to obtain the biosignal information. In addition, the information processing device 1 obtains information representing a drug stimulus given to the individual (such as information representing the concentration of an administered drug) while detecting the biosignal.

[0067] Furthermore, the information processing device 1 acquires information representing the behavior of the individual while the biosignal is being obtained (while the drug stimulus is being administered) as teacher information, uses the acquired biosignal information and information representing the drug stimulus as input information for a machine learning model, and trains the machine learning model by using the acquired teacher information.

[0068] By using the machine learning model that has been machine-learned according to this example, the information processing device 1 can execute the following estimation process. That is, the information processing device 1 inputs, as input information, biosignal information obtained by detecting and digitizing a signal based on the activity of brain nerve cells of the above individual or a human individual different from the above individual (called an estimation subject for distinction), using the sensor S of the biosignal sensor unit 16, and information representing appropriately set drug stimulation, into the machine learning model.

[0069] The information processing device 1 then obtains an output of the machine learning model, and this output is expected to be an estimate of the behavior of the subject in response to the drug concentration.

[0070] [Application example 5.] In the examples so far, the information processing device 1 has performed machine learning processing using biosignal information obtained from an individual and information representing the stimuli (sensory stimuli) given to that individual as input information for a machine learning model, but this embodiment is not limited to this. For example, the machine learning processing may be performed using biosignal information obtained from an individual and information representing the behavior of that individual as input information for a machine learning model, and information representing the stimuli (sensory stimuli) given to the individual while the biosignal information was being obtained as teacher information.

[0071] By using the machine learning model that has been machine-learned in this example, the information processing device 1 becomes able to estimate the sensory stimulus given to the target animal, which is the individual (or an individual different from the individual), based on information representing the behavior. EXAMPLES

[0072] In order to measure the accuracy of mouse behavior estimation by the information processing device 1 of this embodiment, the following experiment was carried out.

[0073] In this experiment, a plurality of sensors S (306 pieces) provided in the biosignal sensor unit 16 of the information processing device 1 are invasively connected to different brain nerve cells of a mouse, which is a target animal, and the information processing device 1 is made to sequentially acquire information representing the behavior of the mouse as teacher information while giving audio information and a stimulus by a spout lever to the mouse. As in the example already described, the information processing device 1 sequentially acquires information representing the behavior of the mouse as teacher information based on images captured by a camera C or the like connected to the input / output unit 15. In this example, the movement trajectory (two-dimensional movement) of the nose as seen from the front of the mouse is used, and 56-dimensional vector information is obtained for 28 body parts. In addition, here, 11-dimensional (about 20%) components are extracted from the 56-dimensional vector information by principal component analysis to generate vector information that serves as teacher information.

[0074] While the mouse connected to the sensor S is given different audio signal stimuli at different times (and while the mouse is given intermittent stimuli via the spout lever in response to its voluntary movements), the information processing device 1 acquires the bio-signals detected sequentially by the sensor S of the bio-signal sensor unit 16.

[0075] The information processing device 1 also inputs, as input information, time series information representing the applied stimulation and corrected biosignal information obtained by selecting N (N is an integer between 0 and 306) dimensions from the biosignal information (306-dimensional vector information) obtained by digitizing the biosignals detected sequentially and setting the elements of the selected biosignal information to "0" (i.e., eliminating the input) into a machine learning model, which is a reservoir network.

[0076] The information processing device 1 sequentially obtains the output of the machine learning model for the above input information, and uses the obtained output of the machine learning model and the sequentially obtained teacher information to update the weights of the machine learning model so as to reduce the difference between them (backpropagation processing).

[0077] This machine learning process is run for a specified period of time, and the machine learning model is then trained on the model while changing N (for example, by setting N = 0, 10, 100, 306) to obtain a machine learning model corresponding to each N.

[0078] Thereafter, the information processing device 1 is made to perform an estimation process as follows. In this estimation process, while the mouse connected to the sensor S is given stimuli of different sound signals at different times (and while the mouse is given intermittent stimuli by the spout lever due to its voluntary behavior), the information processing device 1 acquires biosignals detected sequentially by the sensor S of the biosignal sensor unit 16. The information processing device 1 then inputs, as input information, time-series information representing the given stimuli and corrected biosignal information obtained by digitizing the sequentially detected biosignals (306-dimensional vector information) by selecting N (N is an integer between 0 and 306) dimensions and setting the elements of the selected biosignal information to "0" (i.e., eliminating the input) into the machine learning model corresponding to N obtained by the machine learning process.

[0079] At this time, the information processing device 1 sequentially acquires information representing the mouse's behavior as teacher information based on images captured by a camera C or the like connected to the input / output unit 15. In this example, the movement trajectory (two-dimensional movement) of the mouse's nose as seen from the front is used, and 56-dimensional vector information is obtained for 28 body parts. Here, principal component analysis is used to extract 11-dimensional (approximately 20%) components from the 56-dimensional vector information to generate vector information to be used as comparison information.

[0080] The information processing device 1 sequentially obtains the output of the machine learning model for the above input information, and outputs the obtained output of the machine learning model as information representing the behavior of the mouse.

[0081] The difference between this output and the vector information used as the comparison information was evaluated as follows.

[0082] Figure 3 shows the root mean square error (RMSE) between the output of the machine learning model during the estimation process and the vector information used as comparison information when N is changed. Note that when N = 10 and when N = 100, the initial values ​​(random states) of the reservoir layer were changed multiple times to obtain results.

[0083] As can be seen from the results shown in FIG. 3, the higher the dimension of the inputted biosignal information (that is, the resolution of the inputted biosignal), the more accurate the behavior prediction becomes. [Explanation of symbols]

[0084] 1 information processing device, 11 control unit, 12 memory unit, 13 operation unit, 14 display unit, 15 input / output unit, 16 biosignal sensor unit, 30 machine learning processing unit, 31 signal receiving unit, 32 teacher information receiving unit, 33 machine learning execution unit, 40 estimation processing unit, 41 signal receiving unit, 42 estimation execution unit, 43 output processing unit.

Claims

1. an input means for receiving input of biosignal information representing a biosignal output from a nerve cell of a target animal and information representing a behavior or neural activity of the target animal; an estimation means for executing a predetermined estimation process using a machine learning model that has been machine-learned to learn the relationship between the biological signal information received by the input means and information representing the behavior or neural activity of the target animal that is input together with the input information; and an output means for outputting a result of the estimation process by the estimation means.

2. An input means for receiving an input of information including biosignal information representing a biosignal of a target animal and stimulus information representing the content of a stimulus to the target animal; an estimation means for acquiring estimated information representing the behavior or neural activity of the target animal by inputting input information including information related to the signal received by the input means into a machine learning model that uses the same type of input information as the biological signal information and stimulation information received by the input means and has learned the relationship between the input information and output information representing the behavior or neural activity of the target animal through machine learning; and an output unit that outputs the estimated information acquired by the estimation unit.

3. 3. The information processing device according to claim 2, An information processing device, wherein the biological signal information is biological signal information acquired after the target animal receives the stimulus represented by the stimulus information.

4. 3. The information processing device according to claim 1, The biological signal information received by the input means is information representing a biological signal that changes in a time series, the estimation means sequentially receives input information including biosignal information related to a biosignal that changes in a time series of the same type as the signal received by the input means, uses a machine learning model that has learned the relationship between the input information and output information that represents the behavior or neural activity of the target animal by machine learning, and inputs the input information including biosignal information related to the biosignal that changes in a time series received by the input means to the machine learning model to obtain estimated information that represents the behavior or neural activity of the target animal; The information processing device, wherein the machine learning model includes a recursive neural network.

5. 3. The information processing device according to claim 2, the biological signal received by the input means is a signal that changes in a time series and is a biological signal output by a nerve cell of the target animal; the estimation means uses a machine learning model that has been machine-learned to learn the relationship between input information including biosignal information related to a time-series changing biosignal output by the same type of neuron as the neuron that outputs the biosignal related to the biosignal information received by the input means as the same type of biosignal as the biosignal received by the input means, and output information representing the behavior of the target animal, and inputs input information including biosignal information related to the time-series changing biosignal received by the input means into the machine learning model to obtain estimated information representing the behavior of the target animal; The information processing device, wherein the machine learning model includes a recursive neural network.

6. an input means for receiving input of biosignal information representing a biosignal output from a nerve cell of a target animal and teacher information representing a behavior or neural activity of the target animal while receiving the input of the biosignal information; a machine learning processing means for inputting input information including biosignal information related to a biosignal received by the input means into a machine learning model having an input layer, an intermediate layer, and an output layer, capable of updating weight information between at least the intermediate layer and the output layer, and updating weight information between at least the intermediate layer and the output layer of the machine learning model based on a difference between information output by the machine learning model and teacher information representing the received behavior or neural activity of a target animal, thereby bringing the relationship between the input information and information representing the behavior or neural activity of the target animal into a state of machine learning; Including, An information processing device in which the machine learning model is used to estimate information representing the behavior or neural activity of the target animal.

Images