Reservoir

Abstract

The objective is to provide a reservoir system that enables timely predictions. [Solution] The reservoir system according to this embodiment includes a plurality of sensors and a reservoir, each of the plurality of sensors is configured to send a time-division signal obtained by time-dividing the sensed time-series signal at regular intervals to the reservoir as a sensor signal at regular intervals, the reservoir nonlinearly transforms the sensor signals from each of the plurality of sensors, and the reservoir is configured to output an output signal as future prediction data before the detection of the time-series signal at each of the plurality of sensors is completed.

Description

Technical Field

[0001] The present disclosure relates to a reservoir system. Background Art

[0002] A neuromorphic device is an element that mimics the human brain using a neural network. A neuromorphic device artificially mimics the relationship between neurons and synapses in the human brain.

[0003] For example, a neuromorphic device has nodes arranged in layers (neurons in the brain) and transmission means connecting these nodes (synapses in the brain). In a neuromorphic device, the transmission means (synapses) perform learning to increase the accuracy of correct answers to problems. Learning is the process of finding knowledge that may be useful in the future from information, and in neuromorphic devices, weighting is applied to input data.

[0004] As one type of neural network, a recurrent neural network is known. A recurrent neural network can handle non-linear time-series data. Non-linear time-series data is data whose values change over the passage of time, and stock prices are one example thereof. A recurrent neural network can perform data processing that includes past information by returning processing results from neurons in a subsequent layer back to neurons in a preceding layer.

[0005] Reservoir computing is one means for implementing a recurrent neural network. Reservoir computing performs recursive processing by causing signals to interact based on internal connections. For example, Patent Document 1 discloses an information processing apparatus including a reservoir. Prior Art Documents Patent Documents

[0006] Patent Document 1 International Publication No. 2022 / 024167 [Overview of the project] [Problems that the invention aims to solve]

[0007] Neural networks, including reservoir computing, predict an estimated solution based on the input signal after the input signal has been fully processed. However, waiting for the signal input to complete takes time to predict the estimated solution. This time lag hinders timely prediction. While this time lag can be shortened by using advanced computers to speed up computation, systems using advanced computers lack versatility.

[0008] This disclosure is made in view of the above circumstances and provides a reservoir system that enables timely prediction. [Means for solving the problem]

[0009] A reservoir system according to the first embodiment includes a plurality of sensors and a reservoir. Each of the plurality of sensors is configured to sense a time-series signal and send a time-series sensor signal based on the time-series signal to the reservoir. The reservoir performs a nonlinear transformation of the sensor signals from each of the plurality of sensors. The reservoir is configured to output an output signal as future prediction data before the input of the time-series sensor signals to the reservoir is completed. [Effects of the Invention]

[0010] The reservoir system according to the above embodiment enables timely prediction. [Brief explanation of the drawing]

[0011] [Figure 1] This is a schematic diagram of the reservoir system according to the first embodiment. [Figure 2] This is a diagram illustrating the operation of the reservoir system according to the first embodiment. [Figure 3] This is a diagram illustrating the operation of the reservoir system according to the first embodiment. [Figure 4] This is a schematic diagram of the reservoir system according to the second embodiment. [Modes for carrying out the invention]

[0012] The following description of this embodiment will be given in detail with reference to the drawings as appropriate. The drawings used in the following description may be enlarged for convenience to make the features of this disclosure easier to understand, and the specific configuration of each component may differ from the actual one. The configurations etc. exemplified in the following description are examples, and this disclosure is not limited to them, and can be modified as appropriate to achieve the effects of this disclosure.

[0013] "First Embodiment" Figure 1 is a block diagram of the reservoir system 100 according to this embodiment. The reservoir system 100 comprises a plurality of sensors 1 and a reservoir 2.

[0014] Each of the multiple sensors 1 is connected to a reservoir 2. Each of the sensors 1 senses a time-series signal S1 as an external signal. Each of the sensors 1 is configured to send a time-series sensor signal S2, based on the time-series signal S1, to the reservoir 2.

[0015] The type of sensor 1 is not particularly limited. For example, an image detection sensor, an accelerometer, a temperature sensor, or an inertial sensor can be used as sensor 1. An image detection sensor may, for example, be capable of representing a predetermined point in an image (for example, a red area) using coordinates. An accelerometer may, for example, be attachable to a finger, arm, etc., and capable of detecting arm movement by acceleration. Each of the sensors in sensor 1 may be of the same type or different types.

[0016] The time-series signal S1 is a signal containing information of a measurement object that changes over time. For example, when the sensor 1 is an acceleration sensor, the time-series signal S1 indicates a time change of acceleration in a certain direction. For example, when the sensor 1 is an image detection sensor, the time-series signal S1 indicates a time change of coordinates of a measurement point.

[0017] The sensor 1 may output the time-series signal S1 as it is as a sensor signal S2. In this case, the time-series signal S1 matches the sensor signal S2. When the time-series signal S1 is directly used as the sensor signal S2, all information contained in the time-series signal S1 can be utilized without omission. Further, there is no time lag for converting the time-series signal S1 into the sensor signal S2, and prediction by the reservoir system 100 can be performed in a timely manner.

[0018] Further, the sensor 1 may convert the time-series signal S1 and output the converted signal as the sensor signal S2. For example, the sensor 1 may use, as the sensor signal S2, a time-division signal obtained by time-dividing the time-series signal S1 at fixed time intervals. Time-dividing the time-series signal S1 can reduce the computational load on the reservoir 2.

[0019] For example, the sensor 1 may include a clock and a switch. The clock switches the switch between ON and OFF based on a clock signal. The cycle of the clock signal may be, for example, 1 kHz or 1 MHz. The sensor 1 outputs the sensor signal S2 at fixed cycles based on the clock signal. When the sensor 1 operates based on the clock signal, the time-series signal S1 is converted into the sensor signal S2.

[0020] Further, the sensor 1 may further include a signal holding unit that holds a signal for a fixed period. The signal holding unit holds the signal sensed by the sensor 1 while the switch is switched by the clock signal. By holding signals input during a fixed cycle in the signal holding unit, the time-series signals S1 sensed during the fixed cycle can be collectively output as the sensor signal S2. The signal holding unit is, for example, a capacitor.

[0021] For example, when each of the sensors 1 is a different sensor, the operation cycle of each sensor may differ. For example, depending on the type of the sensor 1, the operation cycle may be faster than that of other types of sensors. The use of a clock makes it possible to synchronize the cycle of the sensor signals S2 input from each of the sensors 1. Synchronizing the cycle of the sensor signals S2 can prevent a signal dependent on one sensor 1 (for example, a signal with an early operation cycle) from being input to the reservoir 2.

[0022] The reservoir 2 is connected to each of the plurality of sensors 1. The sensor signals S2 input from the plurality of sensors 1 to the reservoir 2 are subjected to non-linear conversion within the reservoir 2.

[0023] The reservoir 2 may be implemented as software, hardware, or a combination of software and hardware.

[0024] The software is a program installed in a computer. When the reservoir 2 is implemented as software, the reservoir 2 may be a microcontroller unit (MCU) including a central processing unit (CPU) or a microprocessor unit (MPU). For example, the reservoir 2 includes a memory that stores a reservoir computing program and a processor that executes the reservoir computing program.

[0025] Hardware consists of a combination of real-world elements or circuits. Hardware reservoirs are called physical reservoirs. Physical reservoirs consist of a combination of real-world elements or circuits and realize the concept of reservoir computing with real-world elements or circuits. For example, a physical reservoir may be a physical circuit. A physical reservoir may be, for example, an optical element that utilizes light, an electrical element that utilizes electricity, a magnetic element that utilizes magnetism, or a mechanical element that utilizes vibration. If the physical reservoir has an analog interface, it may have an analog-to-digital converter that converts the output from the physical reservoir to digital. Alternatively, the physical reservoir may be implemented on a PLD (Programmable Logic device) such as an FPGA (Field-Programmable Gate Array).

[0026] The reservoir 2 comprises, for example, a reservoir layer 21 and an output layer 22. The reservoir layer 21 is connected to multiple sensors 1. The reservoir layer 21 is also connected to the output layer 22.

[0027] The reservoir layer 21 stores the sensor signal S2 and converts it into another signal. The reservoir layer 21 converts the time-series sensor signal S2 sequentially without waiting for all of the time-series sensor signal S2 to be input. The reservoir layer 21 converts the data of the time-series sensor signal S2 that has been input to the reservoir 2.

[0028] The reservoir layer 21 has multiple nodes n. Each node n corresponds to a neuron in a neural circuit, and the connections between nodes n correspond to synapses in a neural circuit. Each node n performs a nonlinear transformation of the input and outputs the result. The number of nodes n is not limited. Each node n may be fully connected to multiple sensors 1. Each node n may also be fully connected to the output layer 22.

[0029] Each node n is connected to one or more other nodes n. Each node n may be connected to all other nodes n in the reservoir layer 21, or to some of the nodes n in the reservoir layer 21. The connections between nodes n are random.

[0030] The connections between nodes n include recursive connections. A recursive connection is a connection where the output returns to the input. For example, a signal output from one node n at time t may return to the node n that output the signal at time t+1 or later. This occurs because the signal output from one node n propagates through other nodes n and returns to the original node n. Such a connection relationship between nodes n, where the output from one node n is input again via other nodes n, is called a recursive connection. Reservoir 2 becomes capable of inference based on past information.

[0031] A coupling coefficient, representing the coupling weight, is set between each node n. A signal input to node n propagates between nodes n. The signal propagated to a node n is nonlinearly transformed by the activation function of node n, and then multiplied according to the coupling coefficient before propagating to the next node n. The coupling coefficient between nodes n can be arbitrarily set within the range of -1.0 to +1.0. The coupling coefficient between nodes n can be set, for example, by random numbers.

[0032] Signal exchange between nodes n may use analog signals. Analog signals can handle timely information, enabling timely predictions by the reservoir system 100. Furthermore, if there is no need to convert analog signals to digital signals, the overall power consumption of the reservoir system 100 can be reduced.

[0033] Furthermore, signal exchange between nodes n may be controlled using a clock. For example, each node n may have a clock and a switch. Each node n sends a signal to other nodes at a fixed timing based on the clock signal. As each node n operates based on the clock signal, the time-series signal S1 is converted into a sensor signal S2. Each node n may further have a signal holding unit that holds the signal for a certain period of time. The signal holding unit holds the signal received by node n while the switch is being switched by the clock signal. By holding the signals input during a certain period in the signal holding unit, the signals sensed during that period can be sent together to the next node n. By controlling the signal exchange between nodes n with a clock, the timing of signal propagation between each node n can be synchronized.

[0034] The output layer 22 is connected to the reservoir layer 21. A coupling coefficient (weight) is set between each node n of the output layer 22 and the reservoir layer 21. This coupling coefficient is updated and optimized during training of the reservoir system 100. During inference using the reservoir system 100, this coupling coefficient is fixed according to the training results.

[0035] The output layer 22 applies weights to the output signal S3 from the reservoir layer 21. Applying weights corresponds to multiplying the output signal S3 by the coupling coefficient. The processing of the output layer 22 may be performed as a software calculation, or it may be performed using a hardware wafer, such as a multiply-accumulate circuit.

[0036] The output layer 22 outputs an output signal S4 based on the output signal S3 from the reservoir layer 21. The output signal S3 is a conversion of the sensor signal S2 input to the reservoir 2 and contains some information about the signal input to the reservoir 2. The output signal S4 is output based on the signal input to the reservoir 2.

[0037] The output layer 22 outputs the output signal S4 before the input of the time-series sensor signal S2 to the reservoir 2 is complete. In other words, the timing of the output signal S4 output is earlier than the timing of the completion of the input of the sensor signal S2. The output layer 22 outputs the output signal S4 while the input of the sensor signal S2 is still in progress. The output layer 22 outputs the output signal S4 using past data from the time-series sensor signal S2 that has been input to the reservoir 2 up to the time of the output signal S4 output. In other words, the reservoir 2 can predict future data that has not yet been input based on past data and output the output signal S4.

[0038] For example, if the output layer 22 is a soft wafer, the output layer 22 is programmed to output an inference result based on the current information without waiting for the sensor signal S2 to be input to the reservoir 2, even before the input of the sensor signal S2 to the reservoir 2 is complete.

[0039] For example, if the output layer 22 is hardware, the output layer 22 has a signal propagation path that does not involve capacitors or the like that store all the information of the sensor signal S2 until the input to the reservoir 2 is complete. By using this signal propagation path, the output layer 22 outputs an inference result based on the current information as an output signal S4 even before the input of the sensor signal S2 to the reservoir 2 is complete.

[0040] The output signal S4 may change over time. The output signal S4 may be, for example, an analog signal. The output signal S4 may also be a digital signal obtained by digitally converting an analog signal. The output signal S4 may be output in multiple parts, mainly over time. For example, if the sensor signal S2 is a time-division signal, the number of times the output signal S4 is output may be less than or equal to the number of sampling times the time-division sensor signal S2 is input to the reservoir 2.

[0041] The output signal S4 is output from the output unit in a format that the user can recognize. The output unit may be an image display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display. The output unit may also be an interface for connecting to an image display device. In this case, the output unit generates a video signal for displaying image data and outputs the video signal to the image display device connected to it. The output unit may also be a device that outputs sound, such as a speaker. The output unit may also be an interface for connecting to an audio output device such as a speaker or headphones.

[0042] Next, the operation of the reservoir system 100 according to the first embodiment will be described. The reservoir system 100 is capable of learning and inference.

[0043] First, let's explain how to train the reservoir system 100. When training the reservoir system 100, the output signal S4 is compared with the training signal, and the weights (coupling coefficients) applied to the output signal S3 are adjusted so that the degree of agreement (mutual information) between the output signal S4 and the training signal is high.

[0044] The teacher signal may be stored in a memory unit, for example. The memory unit is configured using a storage medium such as a magnetic hard disk drive or a semiconductor memory device. The memory unit may also be a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor memory devices (e.g., SSDs: Solid State Drives), as well as storage devices such as hard disks and semiconductor memory devices built into computer systems.

[0045] Let's explain with a concrete example. For instance, consider a system where sensor 1 is an image detection sensor and the reservoir system 100 predicts the number drawn by the finger based on the finger's movement. For example, a red finger cot is attached to the finger, and the image detection sensor detects the red point in the image.

[0046] Figure 2 is a diagram illustrating the operation of the reservoir system 100 according to the first embodiment. As shown in Figure 2, when a finger draws the number "2", the movement of the finger can be converted into a change in the coordinates of the X and Y axes. Each of these X-axis and Y-axis movements becomes a time-series signal S1 detected by sensor 1.

[0047] In this case, the correct answer for the reservoir system 100 is to output an output signal S4 of "2". In other words, the solution "2" becomes the training signal. If the reservoir system 100 does not output the solution "2", the weight (coupling coefficient) applied to the output signal S3 is adjusted. This process is performed for various numbers and various finger movements of subjects to optimize the weight (coupling coefficient) applied to the output signal S3.

[0048] Let's explain with another specific example. For instance, consider a system where sensor 1 is an accelerometer and reservoir system 100 predicts the shape of the fingers. For example, an accelerometer is installed on each of the five fingers, and the accelerometer detects the movement of the fingers.

[0049] For example, the movement of the fingers differs when clenching them compared to spreading them, and the signals detected by the accelerometer are also different. For instance, when a finger is clenched, the reservoir system 100 should output an output signal S4 indicating "clenching fingers" based on the time-series signal from the accelerometer. In other words, the "clenching fingers" gesture becomes the training signal. If the reservoir system 100 does not output the "clenching fingers" gesture, the weight (coupling coefficient) applied to the output signal S3 is adjusted. This process is performed for various numbers and the finger movements of various subjects to optimize the weight (coupling coefficient) applied to the output signal S3.

[0050] Next, the inference process of the reservoir system 100 will be explained. When the reservoir system 100 performs inference, sensor 1 senses an external signal as a time-series signal S1. The time-series signal S1 becomes sensor signal S2 and is input to reservoir 2. Within reservoir 2, sensor signal S2 becomes output signal S3 and weights are applied to it. The weights applied at this time are the weights determined during training. Then, the reservoir system 100 outputs output signal S4 based on output signal S3. Output signal S4 is output as future prediction data even before the input of the time-series sensor signal S2 to reservoir 2 is complete.

[0051] Let's explain with a specific example. For instance, consider a case where sensor 1 is an image detection sensor, and the reservoir system 100 is a system that predicts the number drawn by the finger based on the finger's movement.

[0052] Figure 3 is a diagram illustrating the operation of the reservoir system 100 according to the first embodiment. The movement of the finger when drawing a number is input to the reservoir system 100 as a time-series signal S1. The time-series signal S1 is a signal that includes all information from the beginning to the end of drawing the number. The reservoir system 100 starts prediction as soon as the input of the time-series signal S1 begins.

[0053] For example, at frame 10, the numbers have just begun to be written, and the input of the time-series signal S1 is not yet complete. For example, at frame 20, the numbers are still being written, and the input of the time-series signal S1 is not yet complete. For example, the input of the time-series signal S1 is only completed at frame 42.

[0054] The reservoir system 100 outputs an estimated solution at each stage. For example, at frame 5, it outputs an estimated solution that there is a possibility that it is drawing "7" or "8". For example, at frame 20, it outputs an estimated solution that there is a possibility that it is drawing "7". For example, at frame 42, it outputs an estimated solution that there is a high probability that it was drawing "2".

[0055] For example, when drawing the number "2", there is a characteristic bend just before drawing the last horizontal line. This type of movement is characteristic of "2" or "3", and at approximately 25 frames, the reservoir system 100 outputs an estimated solution indicating that "2" has the highest probability.

[0056] In the first embodiment, the reservoir system 100 predicted the finger movements approximately 20 frames before the numerical values ​​were finished being drawn in 42 frames. The reservoir system 100 in the first embodiment can make timely predictions and reduce the time lag.

[0057] Here, the reservoir system 100 outputs an inappropriate estimated solution up to frame 20. This is something that happens even in normal human judgment and is not a problem. For example, the same thing happens when a person predicts the number to be drawn while the number "2" is still being drawn. Even if the prediction is incorrect, making a timely prediction can speed up subsequent decision-making.

[0058] Here, we have provided a specific example of a system that predicts hand movements using an image detection sensor, but similar timely predictions can be made with other sensors as well.

[0059] For example, even in a system that uses an accelerometer attached to a finger or arm to predict the shape of the fingers, the shape of the fingers can be predicted from information about changes in acceleration of the finger or arm even before the actual action of "clenching" or "spreading" is completed. For instance, if the shape of the fingers can be predicted, in a game of rock-paper-scissors, it would be possible to predict what hand shape the opponent will make before they actually make the move.

[0060] Here, we have shown an example where learning and inference are performed at different times, but learning and inference can also be performed in parallel. For example, learning can be performed online. In online learning, the coupling coefficients are updated each time a data point is input. By performing online learning, the accuracy of the estimated solution can be improved even when the response changes over time.

[0061] "Second Embodiment" Figure 4 is a schematic diagram of a reservoir system 101 according to the second embodiment. The reservoir system 101 comprises a plurality of sensors 1, a reservoir 2, an input detection unit 3, and an output detection unit 4. The reservoir system 101 differs from the reservoir system 100 in that it has an input detection unit 3 and an output detection unit 4. In the reservoir system 101, components similar to those in the reservoir system 100 are denoted by the same reference numerals and their descriptions are omitted.

[0062] The input detection unit 3 is connected to each of the multiple sensors 1 and to the reservoir 2. The input detection unit 3 is configured to monitor the sensor signals S2 and determine the starting point of the signal from the sensor signals S2 that will be used for predicting the prediction data.

[0063] The input detection unit 3 may be software, hardware, or a combination of software and hardware.

[0064] The input detection unit 3 may be, for example, a microcontroller unit (MCU) or microprocessor unit (MPU) including a central processing unit (CPU), or it may be an input detection circuit.

[0065] The input detection unit 3 determines the point of change as the starting point when, for example, the sensor signal S2 from sensor 1 changes from a certain state. The degree of change from the certain state can be freely designed according to the task, application, etc., given to the reservoir system 101. For example, the input detection unit 3 may determine the starting point as the point when the amount of change in the sensor signal S2 exceeds a certain threshold, or it may determine the starting point as the point when the sensor signal S2 exceeds a certain threshold.

[0066] The output detection unit 4 is connected to the reservoir 2. The output detection unit 4 is configured to monitor the output signal S4 and determine the endpoint of the signal from the sensor signal S2 that will be used for predicting the predicted data.

[0067] The output detection unit 4 may be software, hardware, or a combination of software and hardware.

[0068] The output detection unit 4 may be, for example, a microcontroller unit (MCU) or microprocessor unit (MPU) including a central processing unit (CPU), or it may be an output detection circuit.

[0069] The output detection unit 4 determines, for example, the starting point when the output signal S4 changes from a certain state. The degree of change from the certain state can be freely designed according to the task, application, etc., given to the reservoir system 101. For example, the output detection unit 4 may determine the starting point when the amount of change in the output signal S4 exceeds a certain threshold, or it may determine the starting point when the output signal S4 exceeds a certain threshold.

[0070] The output detection unit 4 may also be located between each of the multiple sensors 1 and the reservoir 2. The output detection unit 4 may determine the point of change as the termination point when the sensor signal S2 changes from a certain state. For example, the output detection unit 4 may determine the termination point as the point when the amount of change in the sensor signal S2 falls below a certain amount.

[0071] The reservoir system 101 according to the second embodiment can perform timely predictions, similar to the reservoir system 100 according to the first embodiment. Furthermore, since the start and end points of the prediction can be set, the reservoir system 101 does not need to perform unnecessary calculations, and the power consumption of the reservoir system 101 can be reduced.

[0072] In the reservoir system 101, an example is shown that has both an input detection unit 3 and an output detection unit 4, but either one alone may be used. Even just setting a start point or an end point can significantly reduce the computational load.

[0073] Although several embodiments have been described in detail above with reference to the drawings, the configurations and combinations thereof in each embodiment are merely examples, and additions, omissions, substitutions, and other modifications to the configurations are possible without departing from the spirit of this disclosure. [Explanation of symbols]

[0074] 1 sensor 2 Reservoirs 3. Input detection unit 4. Output detection unit 21 Reservoir layer 22 Output Layers 100, 101 Reservoir System n nodes S1 Time-series signal S2 Sensor Signal S3, S4 output signals

Claims

1. It has multiple sensors and a reservoir, Each of the aforementioned multiple sensors is configured to send a time-division signal, obtained by time-dividing the detected time-series signal at regular intervals, to the reservoir at regular intervals as a sensor signal. The reservoir performs a nonlinear transformation of the sensor signals from each of the plurality of sensors. A reservoir system in which the reservoir is configured to output an output signal as future predictive data before the detection of the time-series signals at each of the multiple sensors is completed.

2. The reservoir system according to claim 1, wherein the output signal is determined using past data from the sensor signals that have been input to the reservoir up to the time the output signal is output.

3. The reservoir outputs the output signal multiple times as time progresses. The reservoir system according to claim 1, wherein the number of times the output signal is output from the reservoir is less than or equal to the number of sampling times the time-divided time-series signal is input to the reservoir.

4. The aforementioned reservoir comprises a reservoir layer and an output layer, The reservoir layer converts the sensor signal that is input at regular intervals, The reservoir system according to claim 1, wherein the output layer applies a weight to the output from the reservoir layer and outputs the output signal based on the signal input to the reservoir.

5. The reservoir system according to claim 1, wherein the output signal is an analog signal.

6. The reservoir has multiple nodes, The reservoir system according to claim 1, wherein the signal exchange between the plurality of nodes is performed using analog signals.

7. The reservoir has multiple nodes, The reservoir system according to claim 1, wherein each of the plurality of nodes sends a signal to the other nodes at a predetermined period.

8. The reservoir system according to claim 7, wherein the period for sending a signal to another node in each of the plurality of nodes matches the period for the sensor signal sent to the reservoir.

9. The reservoir system according to claim 1, wherein at least one of the plurality of sensors is an image detection sensor.

10. The reservoir system according to claim 1, wherein at least one of the plurality of sensors is an acceleration sensor.

11. It further includes an input detection unit, The input detection unit is connected to each of the plurality of sensors and the reservoir, The reservoir system according to claim 1, wherein the input detection unit is configured to monitor the sensor signal and determine the starting point of the signal from the sensor signal to be used for predicting the prediction data.

12. It is further equipped with an output detection unit, The output detection unit is connected to the reservoir, The reservoir system according to claim 1, wherein the output detection unit is configured to monitor the output signal and determine the termination point of the signal used for predicting the prediction data from among the sensor signals.

13. The aforementioned plurality of sensors are acceleration sensors that can be worn on a finger or arm. The reservoir system according to claim 1, wherein the reservoir is configured to predict hand movements as the output signal.

14. The aforementioned plurality of sensors are sensors that can indicate a predetermined point in an image using coordinates, The reservoir system according to claim 1, wherein the reservoir is configured to predict hand movements as an output signal using the transition of the coordinates.

Images