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

Abstract

This invention provides an information processing device, system, method, and program that suitably perform various information processing tasks, including robot control, using a quantum reservoir with a small number of qubits (NISQ device). [Solution] In an information processing system in which a first information processing device 10, which is a determination device, and a second information processing device 20, which is a quantum computer, can communicate with each other via a network N, the first information processing device 10 includes a first acquisition unit 12 for acquiring input data, a generation unit 20a for generating reservoir input data to be input to a quantum reservoir 20 having multiple layers of sub-reservoirs from the input data, and an output data generation unit 14 for generating output data by referring to the output result, and the second information processing device 20 includes a second acquisition unit 20b for acquiring the output result of the quantum reservoir to which the reservoir input data has been input.

Description

Technical Field

[0005] ,

[0004] , ,

[0001] The present invention relates to an information processing apparatus, an information processing method, an information processing system, a robot system, and a program.

Background Art

[0002] In recent years, artificial intelligence has been utilized in various fields, and the development of new artificial intelligence technologies has been advanced. As a kind of next-generation artificial intelligence, a machine learning method called reservoir computing has attracted attention. Also, research on robot motion control and various information processing using reservoir computing has attracted attention.

[0003] Reservoir computing is a method based on a recurrent neural network, which is a complex model with non-linearity. Reservoir computing is composed of an input layer, a reservoir layer, and an output layer. After converting time-series input data by the intermediate layer fixed during learning called the reservoir layer, only the readout part is trained by a simple learning method such as linear regression, so fast learning is possible. Therefore, it is expected to be applied in scenarios where data processing and analysis are performed in real time.

[0004] The reservoir layer, which plays an important role in reservoir computing, only needs to be able to generate useful spatio-temporal patterns from input data, so a method of substituting it with a complex physical system has been suggested. In theory, it is possible to create a high-dimensional spatio-temporal space by utilizing the non-linearity of a large degree of freedom of a complex physical system. In that high-dimensional spatio-temporal space, data learning can be easily performed. Such a reservoir is suitable for control of robots and various information processing because it consumes less power and enables high-speed processing.

[0005] However, designing a physical system to be used as a reservoir layer that can achieve high learning accuracy is difficult. This is because, although reservoir calculation methods using soft robotics and tensagrency as reservoirs have been proposed, it is not obvious what kind of physical system to design and how to design it to achieve high accuracy (see, for example, Non-Patent Documents 1 and 2).

[0006] To address these challenges, various physical systems have been used as reservoirs. Recently, however, a method has been proposed that utilizes the dynamics of the qubits (quantum bits) of a complex quantum computer with high degrees of freedom for time-series learning. This is because the quantum dynamics reproduced by a quantum computer with n qubits can utilize the 2n-dimensional quantum space for feature extraction of time-series data, and it is expected that high-precision learning can be achieved by effectively utilizing this.

[0007] While the design of reservoir layers using such quantum computers is still in the research stage, it has recently been proposed to design reservoir layers using the quantum states of qubits (see, for example, Non-Patent Documents 3 and 4). [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] Caluwaerts, K., Despraz, J. Iscen,A., Sabelhaus, AP, Bruce, J., Schrauwen, B., & SunSpiral, V. (2014).Design and control of compliant tensegrity robots through simulation and hardware validation. JR Soc. Interface 11, 98. [Non-Patent Document 2] Nakajima, K., Hauser, H., Li, T., & Pferfer, R. (2015). Information Processing via physical soft body. Sci.Rep. 5, 10487. [Non-Patent Document 3] Fujii, K., & Nakajima, K. (2017 ). Harnessing disordered-ensemble quantum dynamics for machine learning. Physical Review Applied, 8(2), 024030. [Non-Patent Document 4] Chen, J., Nurdin, HI, &Yamamoto, N. (2020). Temporal information processing on noisy quantum computers.arXiv preprint arXiv: 2001.09498. [Overview of the project] [Problems that the invention aims to solve]

[0009] In the method described in Non-Patent Document 3, learning accuracy is heavily dependent on the performance of the quantum computer. However, existing quantum computers are medium-scale and error-prone devices. Therefore, reservoirs using existing quantum computers produce outputs that differ from those of an ideal quantum computer (an error-free, large-scale qubit-based quantum computer), resulting in the challenge of not being able to learn with high accuracy. Such quantum computers are called NISQ (Noisy Intermediate-Scale Quantum) devices, and it is predicted that the era of NISQ devices will continue until at least around 2030-2040. In order to achieve practical applications such as robot control and various information processing using reservoirs with quantum computers, the development of quantum reservoirs that function with NISQ devices is required.

[0010] The present invention aims to provide a technology that can suitably perform various information processing tasks, including robot control, using a quantum reservoir with a small number of qubits (NISQ device). [Means for solving the problem]

[0011] As a result of diligent research, the inventors have found that errors occurring in NISQ devices originate from the quantum noise of the qubits, and that this noise can be considered a type of quantum dynamics. By deliberately utilizing the qubit noise in the reservoir design, the above objective can be achieved. This invention is based on this finding.

[0012] An information processing device according to one aspect of the present invention includes a first acquisition unit for acquiring input data, a generation unit for generating reservoir input data to be input to a quantum reservoir having multiple layers of sub-reservoirs from the input data, a second acquisition unit for acquiring the output result of the quantum reservoir to which the reservoir input data has been input, and an output data generation unit for generating output data by referring to the output result.

[0013] An information processing device according to one aspect of the present invention includes: a first acquisition unit that acquires training data including input data and label data; a generation unit that generates reservoir input data from the input data to be input to a quantum reservoir having multiple layers of sub-reservoirs; a second acquisition unit that acquires the output result of the quantum reservoir to which the reservoir input data has been input; and a learning unit that trains an output data generation unit that generates output data by referring to the output result, using the label data.

[0014] An information processing device according to one aspect of the present invention comprises a quantum reservoir having multiple layers of sub-reservoirs, a first acquisition unit that acquires reservoir input data generated from time-series data, a second acquisition unit that acquires the output result of the quantum reservoir after inputting the reservoir input data to the quantum reservoir, and an output unit that outputs the output result.

[0015] An information processing system according to one aspect of the present invention is an information processing system including a first information processing device and a second information processing device, wherein the first information processing device includes a first acquisition unit for acquiring input data, a generation unit for generating reservoir input data to be input to a quantum reservoir having multiple layers of sub-reservoirs from the input data, a second acquisition unit for acquiring the output result of the quantum reservoir to which the reservoir input data has been input, and an output data generation unit for generating output data by referring to the output result, wherein the second information processing device includes a quantum reservoir having multiple layers of sub-reservoirs, a third acquisition unit for acquiring reservoir input data generated by the generation unit of the first information processing device, a fourth acquisition unit for acquiring the output result of the quantum reservoir after the reservoir input data has been input to the quantum reservoir, and an output unit for outputting the output result.

[0016] An information processing method according to one aspect of the present invention includes: a first acquisition step in which one or more processors acquire input data; a generation step in which one processor generates reservoir input data from the input data to be input to a quantum reservoir having multiple layers of sub-reservoirs; a second acquisition step in which one acquires the output result of the quantum reservoir to which the reservoir input data has been input; and an output data generation step in which one acquires output data by referring to the output result.

[0017] An information processing method according to one aspect of the present invention includes: a first acquisition step in which one or more processors acquire training data including input data and label data; a generation step in which reservoir input data to be input to a quantum reservoir having multiple layers of sub-reservoirs from the input data; a second acquisition step in which the output result of the quantum reservoir to which the reservoir input data has been input is acquired; and a learning step in which an output data generation unit that generates output data by referring to the output result is trained using the label data.

[0018] An information processing method according to an aspect of the present invention includes: a first acquisition step in which one or more processors acquire reserve input data generated from time-series data; a second acquisition step in which, after inputting the reserve input data into a quantum reservoir having a plurality of hierarchical sub-reservoirs, an output result of the quantum reservoir is acquired; and an output step of outputting the output result.

[0019] A program according to an aspect of the present invention is a program for causing a computer to function as an information processing apparatus, the program causing the computer to execute: a first acquisition step of acquiring input data; a generation step of generating reserve input data to be input into a quantum reservoir having a plurality of hierarchical sub-reservoirs from the input data; a second acquisition step of acquiring an output result of the quantum reservoir into which the reserve input data has been input; and an output data generation step of generating output data by referring to the output result.

[0020] Another program according to an aspect of the present invention is a program for causing a computer to function as an information processing apparatus, the program causing the computer to execute: a first acquisition step of acquiring teacher data including input data and label data; a generation step of generating reserve input data to be input into a quantum reservoir having a plurality of hierarchical sub-reservoirs from the input data; a second acquisition step of acquiring an output result of the quantum reservoir into which the reserve input data has been input; and a learning step of causing an output data generation unit that generates output data by referring to the output result to learn using the label data.

Effect of the Invention

[0021] According to the present invention, information processing that makes use of the advantages of a quantum reservoir using a quantum computer (NISQ device) with a small number of qubits can be suitably performed.

Brief Description of the Drawings

[0022] [[ID=二十一]] [[ID=二十二]] [Figure 1] It is a schematic diagram of a robot system according to an embodiment of the present invention. [Figure 2] It is a conceptual diagram of a reservoir layer. [Figure 3] It is a diagram showing a construction example of a reservoir layer. [Figure 4] It is a schematic diagram of a quantum circuit. [Figure 5] It is a conceptual diagram of a reservoir layer using a quantum computer. [Figure 6] It is a schematic diagram of an information processing system that uses a quantum computer through the cloud.

Embodiments for Carrying Out the Invention

[0023] Hereinafter, embodiments of the present invention will be described based on the drawings. As shown in FIG. 1, the robot system according to the present embodiment includes a robot 40, a determination device 10, and a controller 30, and includes a robot control system that controls the robot 40.

[0024] The robot 40 has a manipulator 42 and a sensor 44. For example, when the robot 40 is a humanoid robot, the arm corresponds to the manipulator 42. The sensor 44 senses the operation of the manipulator 42 and generates a time-series signal. The sensor 44 according to the present embodiment mainly senses the motion of the robot 40 in time series. The sensor 44 may be incorporated in the manipulator 42 or provided at a position different from the manipulator 42.

[0025] For example, when the manipulator 42 is a robot hand that grasps and lifts an object, as the sensor 44, a pressure sensor provided at a portion where the manipulator 42 grasps the object can be mentioned. The pressure sensor preferably can continuously capture and output a change in pressure instead of distinguishing between two values of whether or not an object has contacted.

[0026] When the robot hand grasps an object, the pressure sensor outputs a continuous pressure change corresponding to the grasping force. By transmitting the output value of the pressure sensor to a computer and learning, the robot 40 can grasp a specific object with an appropriate force.

[0027] In addition to the gripping portion of the manipulator 42, a vibration sensor 44 may be attached to analyze the situation based on vibrations when gripping an object. The vibration sensor outputs continuous vibration changes when the manipulator 42 carries an object. By transmitting the output values ​​of the vibration sensor to a computer and learning from them, the robot 40 can correct the posture of the arm that grips the object, thereby enabling stable transport of the object.

[0028] A camera sensor, acting as a sensor 44, may be attached to the manipulator 42 to analyze the surrounding conditions when grasping an object. For example, when several manipulators work together to transport an object in the same location, a camera sensor mounted on one manipulator outputs a time-series image file that identifies the movements of the surrounding manipulators. By sending the output time-series image file to a computer and learning from it, balanced and coordinated work with other manipulators can be achieved.

[0029] Other sensors that can be suitably used include those that provide continuous or intermittent data over time, specifically sound sensors, tactile sensors, and the like.

[0030] The controller 30 controls the manipulator 42 by issuing pre-programmed operation commands to the manipulator 42. The controller 30 may also be integrated into the manipulator 42.

[0031] The manipulator 42 operates according to commands instructed by the controller 30. The sensor 44 outputs a time-series signal in response to the operation of the manipulator 42.

[0032] The determination device 10 includes an acquisition unit 12, a determination unit 14, and a quantum computer 20.

[0033] The acquisition unit 12 acquires time-series signals (sensor data) from the sensor 44 via the network. The network is a wired or wireless communication network, and may be the Internet or a LAN.

[0034] The quantum computer 20 has an input layer, a reservoir layer, and an output layer. The time-series signal acquired by the acquisition unit 12 is input to the input layer of the quantum computer 20, and the signal is learned according to the schematic diagram in Figure 2, which will be described later.

[0035] The determination unit 14 determines the next action it wants the manipulator 42 to perform based on the output of the quantum computer 20, and notifies the controller 30 of the determination result. The controller 30 translates the notified determination result into an action command for the manipulator 42 and controls the manipulator 42. By repeating this cycle, comprehensive control of the manipulator 42's actions is achieved.

[0036] For example, if the manipulator 42, acting as a robotic hand, grasps an object and transports it to a destination corresponding to the type of object, the output of the sensor 44 is input to the quantum computer 20, and the determination unit 14 notifies the controller 30 of the determination result of the type of object based on the output of the quantum computer 20. The controller 30 controls the manipulator 42 to transport the object to the destination corresponding to the notified type. The object may be waste, food such as fruits or vegetables, or any other object.

[0037] The acquisition unit 12 and the determination unit 14 can be configured as a (classical) computer equipped with a CPU, ROM, RAM, communication unit, etc.

[0038] In order to control the complex movements of the manipulator 42 with high precision using this robot control system, it is necessary to design a reservoir layer with high-precision learning capabilities. The reservoir layer according to this embodiment will be explained using the conceptual diagram shown in Figure 2.

[0039] The quantum computer 20 consists of an input layer for inputting time-series data necessary for learning, a reservoir layer for mapping the data into a higher-dimensional quantum space, and an output layer for outputting the learning results.

[0040] The reservoir layer contains n qubits (where n is an integer greater than or equal to 2), and in NISQ devices, it is assumed that the reservoir layer is affected by noise interacting through the external environment. In this embodiment, in order to utilize complex quantum dynamics by using this noise, the qubits in the reservoir layer are first grouped. Each group contains two or more qubits. Hereinafter, these groups will be referred to as sub-reservoirs.

[0041] Next, a specific quantum circuit is designed for each sub-reservoir, depending on the time-series data to be trained. The quantum circuits in each sub-reservoir may be the same or different.

[0042] Next, the designed quantum circuit is used to input the time-series data to be trained into the sub-reservoirs in chronological order, and the quantum circuit is executed. As a result, the training data is mapped into a higher-dimensional quantum space through the individual sub-reservoirs.

[0043] Furthermore, each sub-reservoir receives crosstalk noise from surrounding sub-reservoirs when executing a quantum circuit, so the time-series data is mapped differently to quantum space for each sub-reservoir. This improves the degrees of freedom and nonlinearity of the reservoir layer, and consequently, the output results of each sub-reservoir obtained by observing the execution results of the quantum circuit will also differ.

[0044] Finally, the connection weights of the output layer are trained using linear regression to minimize the error between the output value and the target value.

[0045] In other words, the acquisition unit 12 acquires the time-series data to be learned (training data) and the target value (label data associated with the time-series data), and the determination device 10 is configured to input the time-series data into the input layer of the quantum computer 20, thereby training the output layer to minimize the error between the output value output from the output layer and the target value.

[0046] To verify the effectiveness of the reservoir layer in this embodiment, two conditions must be confirmed using a quantum computer. First, the reservoir system must be nonlinear. Second, the reservoir system must have a property (called fading memory) in which the current input has a greater influence on the current internal state than past inputs or states.

[0047] Empirically, in order to achieve high-performance reservoir computing, the reservoir must possess this nonlinearity and fading memory property. To verify these two properties, the effectiveness of the reservoir layer in this embodiment was verified using a standard benchmark test task called the NARMA (Nonlinear autoregressive moving average) task. Equation (1) shows the NARMA task.

[0048]

number

[0049] Figure 3 shows an example of constructing a quantum reservoir using IBM's 16-Qubit quantum computer (ibmq_16_melbourne) and training it with 100 time-step data for the NARMA task given by equation (1).

[0050] As shown in Figure 3, first, one sub-reservoir is created using two adjacent qubits (for example, the 0th and 14th qubits). Similarly, another sub-reservoir is created using two other adjacent qubits (for example, the 1st and 13th qubits). This process is repeated to create sub-reservoirs, and finally, a reservoir layer is created using multiple connected sub-reservoirs. Figure 3 shows an example of a reservoir layer constructed with five sub-reservoirs.

[0051] In data training for a 100-time step NARMA task, the quantum circuit shown in Figure 4 is used for each sub-reservoir. Data from each time step of the time series to be trained is input, the quantum circuit is computed, and the output is obtained.

[0052] Figure 4 shows an example of a quantum circuit with time step = 2 using 2 qubits. The quantum circuit shown in Figure 4 has data inputs for time step = 1 and time step = 2. Each input contains 5 quantum gates.

[0053] In this quantum circuit, first, an X rotation operation is applied to each qubit, with the input data as the rotation angle. Next, a CNOT gate is applied across the two qubits, a Z rotation operation is applied to the second qubit, and then another CNOT gate is applied to create a strong correlation between the two qubits. However, the rotation angle for the Z rotation operation is also the input data.

[0054] By repeatedly applying such a quantum circuit at each time step, a quantum state dependent on time-series data is generated. It should be noted that while this quantum circuit structure is used in this embodiment, different quantum gates may be used.

[0055] In this way, in the determination device 10, the time-series data acquired by the acquisition unit 12 is divided into data for multiple time steps (time slices), thereby generating reservoir input data. Then, the data or representative value of the data at each time step included in the generated reservoir input data is input to each sub-reservoir via the input layer. Here, the time-series data division process into multiple time steps may be performed by the input layer or by other configurations in the determination device 10.

[0056] Furthermore, as described above, the determination device 10 performs one or more quantum operations on the qubits constituting each sub-reservoir. The information specifying the quantum operation may, for example, be included in the data input to the input layer or reservoir layer.

[0057] Furthermore, the specific unitary transformations (unitary gates) for the two qubits mentioned above, such as the X rotation operation, CNOT gate, and Z rotation operation, are limited in terms of the overall degrees of freedom of unitary transformations. By using these restricted unitary transformations, the number of quantum gate operations is reduced, resulting in a quantum circuit (hardware-efficient ansatz) that can be executed on NISQ devices. In other words, the complexity of the dynamics in the quantum reservoir is not due to the degrees of freedom of the unitary transformations, but rather to the effects of the noise mentioned above. Therefore, even a simple quantum circuit like a restricted unitary transformation can perform the function of a quantum reservoir.

[0058] As shown in Figure 3, in a reservoir layer constructed with five sub-reservoirs, different outputs were obtained even though the same quantum circuit was used for calculations in each sub-reservoir. This is because, when executing the quantum circuit in each sub-reservoir, it was affected by interaction noise with the surrounding sub-reservoirs and other external environments surrounding the qubit.

[0059] In other words, it was demonstrated that the quantum noise of qubits could be utilized in the construction process of the reservoir layer in this embodiment. Furthermore, it was shown that a high-dimensional quantum space with many degrees of freedom was created because the outputs obtained from each sub-reservoir were significantly different.

[0060] Table 1 shows the NMSE (Normalized Mean Squared Error) values ​​of the prediction results for the NARMA task. For comparison, the NMSE values ​​for linear regression are also shown. The NMSE value is a standard method for evaluating prediction accuracy and is calculated using equation (2).

[0061]

number

[0062] [Table 1] As shown in Table 1, it was confirmed that both the mean and standard deviation of the NMSE decreased with increasing the number of sub-reservoirs in the reservoir layer, indicating an improvement in prediction accuracy. From these results, it was confirmed that with two or more sub-reservoirs, the prediction accuracy is several tens of percent better than conventional linear regression learning. Furthermore, with a quantum reservoir using six sub-reservoirs, a prediction accuracy several times higher than the NSME value of linear regression was obtained, demonstrating the superior computational accuracy of the quantum reservoir in this embodiment.

[0063] The robot motion control performance of the robot control system according to this embodiment was verified. An experiment was conducted to determine whether a reservoir could predict the movement of a robot hand in a robot manipulator having a two-fingered robot hand and a robot arm.

[0064] A vacuum-driven soft gripper from piab was used for the two-fingered robotic hand. A 4-axis robotic arm (Dobot Magician, a desktop-compatible linkage mechanism) was used for the robotic arm. The robotic hand's movements were controlled using software called Dobot Studio.

[0065] A film sensor was attached to a two-fingered robotic hand. The film sensor has a negatively charged dome-shaped silicon layer and a positively charged nylon layer, with copper and aluminum electrode layers added to the silicon and nylon layers.

[0066] When the robot hand performs actions such as grasping and releasing an object, pressure is applied to the dome-shaped silicon layer, generating negative and positive frictional charges on the electrodes, and thus producing a voltage.

[0067] Three plastic objects (Object A, Object B, and Object C) with nearly the same weight (approximately 16g) and volume (approximately 27cm³) but different shapes were prepared. Objects A and B are cubes, while Object C is a sphere. Object A is made of ABS synthetic plastic, and Objects B and C are made of PLA synthetic plastic.

[0068] For each object, a two-fingered robotic hand equipped with a film sensor performed a grasping and releasing motion 25 times. The film sensor's voltage changed over time in response to the robotic hand's finger movements, yielding time-series voltage data. Voltage data for both fingers was obtained for each object. The experiment then verified whether a reservoir computation using a quantum computer could identify objects by learning from the time-series voltage data.

[0069] In the reservoir training, we first used two-finger data to embed the data into a 2-Qubit sub-reservoir on IBM's 27-Qubit quantum computer (ibmq_toronto), as shown in Figure 5, and then performed the same quantum circuit execution and measurement as in the NARMA task.

[0070] Using the output results obtained from the measurements, we adjusted the output weights to create a predictive model that could distinguish between three types of objects. Table 2 shows the results of 10-fold cross-validation for the three types of objects.

[0071] [Table 2] As shown in Table 2, the learning accuracy and prediction accuracy for the discrimination of the three types of objects were 0.87 and 0.83, respectively. This demonstrates that the reservoir layer in this embodiment can accurately learn the sensor's time-series data, discriminate objects, and accurately control the manipulator through the controller.

[0072] In the above embodiment, the robot control system may be a system that integrates the determination device 10, the controller 30, and the robot 40, or it may be configured to use the quantum computer 20 via the cloud.

[0073] In other words, the determination device 10 according to this embodiment may be configured in a distributed manner by a determination device (also called determination device 10a) equipped with an acquisition unit 12 and a determination unit 14, and a quantum computer 20 that is communicatively connected to the determination device 10a. In this configuration, the determination device 10a and the quantum computer 20 communicate with each other via their respective communication units, and the acquisition unit 12 also acquires various data from the quantum computer 20.

[0074] Furthermore, in the configuration where the quantum computer 20 is used via the cloud as described above, some of the components of the quantum computer 20 may be provided by the determination device 10a. For example, at least one of the input layer and the output layer may be provided by the determination device 10a instead of the quantum computer 20. For instance, in the configuration where the output layer is provided by the determination device 10a, the learning process of the output layer can be performed without transmitting the coupling weights of the output layer to the quantum computer 20.

[0075] Thus, in a configuration that uses the quantum computer 20 through the cloud described above, the robot control system according to this embodiment may be configured to include a determination device 10a and a quantum computing device equipped with a quantum reservoir.

[0076] Here, the determination device 10a is, as an example, A first acquisition unit (the acquisition unit 12 described above) acquires time-series signals from sensors installed on the robot, A generation unit (for example, the input layer described above) generates reservoir input data from the aforementioned time-series signal to be input to a quantum reservoir (the reservoir layer described above) having multiple layers of sub-reservoirs, A second acquisition unit (the acquisition unit 12 described above) acquires the output result of the quantum reservoir that has received the reservoir input data, and A determination unit (the output layer and determination unit 14 described above) that determines the operation of the robot by referring to the output result, It is equipped with.

[0077] Furthermore, the aforementioned quantum computing device equipped with a quantum reservoir, as an example, A quantum reservoir (the reservoir layer described above) has multiple layers of sub-reservoirs. The system acquires reservoir input data generated from time-series signals obtained from sensors installed on the robot. After inputting the reservoir input data to the quantum reservoir, the output result of the quantum reservoir is obtained. The output result is output to the determination device 10a.

[0078] Furthermore, the determination device 10a that performs the learning process described above is A first acquisition unit (acquisition unit 12 described above) acquires training data including time series data and label data, From the aforementioned time-series data, a generation unit (the input layer described above) generates reservoir input data to be input to a quantum reservoir (the reservoir layer described above) having multiple layers of sub-reservoirs, A second acquisition unit (the acquisition unit 12 described above) acquires the output result of the quantum reservoir that has received the reservoir input data, and Equipped with, A determination unit (the output layer and determination unit 14 described above) that determines the robot's operation by referring to the output results is trained using the label data.

[0079] <Additional Note 1> In the above explanation, robot control was given as an example of inference and learning processing using a quantum reservoir with multiple layers of sub-reservoirs. However, as can be seen from the above explanation, processing using a quantum reservoir with multiple layers of sub-reservoirs is not limited to robot control, but can be applied to various other fields. Examples include medical care, mobility, collection and delivery, and sorting work.

[0080] For example, the acquisition unit 12 may acquire medical data, such as the subject's electroencephalogram (EEG), pulse wave, and respiratory rate, as time-series data, and the determination unit 14 may perform inferences regarding the subject's physical condition. The time-series data may also include the subject's blood glucose level, drug concentration, dosage, and blood oxygen saturation. The determination device 10 may output the inference results from the determination unit 14 in the form of advice to the subject, such as images or audio. In such a configuration, the output layer can be trained using training data that includes the subject's vital data such as electroencephalogram (EEG), pulse wave, respiratory rate, blood glucose level, drug concentration, dosage, and blood oxygen saturation, as well as label data related to the subject's physical condition associated with the vital data.

[0081] Furthermore, in the configuration described above, the acquisition unit 12 and the output unit that outputs the inference results from the determination unit 14 may be provided in separate devices. For example, vital data as time-series data may be acquired by the acquisition unit 12 provided in a testing device installed in a hospital in one country, the time-series data may be input to a quantum reservoir installed in another country, and an image including advice generated by the determination unit 14 based on the output results from the quantum reservoir may be displayed on a display terminal outside the hospital.

[0082] As another example, the acquisition unit 12 may acquire sensing data from various sensors installed on the target vehicle as time-series data, and the determination unit 14 may perform driving control of the target vehicle. In such a configuration, the output layer can be trained using training data that includes sensing data from various sensors installed on the target vehicle and label data related to driving control associated with the sensing data.

[0083] Furthermore, the acquisition unit 12 may acquire, as time-series data, not only sensing data from various sensors installed on the target vehicle, but also data from other vehicles traveling around the target vehicle and cameras installed near the road. Thus, the time-series data acquired by the acquisition unit 12 may include time-series data from separate devices and equipment. The generation unit described above may be configured to generate reservoir input data based on or after applying integrated processing to such sensing data. The determination unit 14 may be configured to perform driving control of the target vehicle based on the output result of the quantum reservoir that has received such reservoir input data. In such a configuration, the types and number of channels of sensing data will increase. Accordingly, it is preferable that the quantum reservoir has a number of qubits corresponding to the types and number of channels of sensing data. Even when using a quantum reservoir with such a large number of qubits, the invention described herein is useful.

[0084] Therefore, the embodiments described herein can also be expressed as follows:

[0085] (Aspect A1) A first acquisition unit that acquires input data, A generation unit generates reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition unit that acquires the output result (measurement result) of the quantum reservoir to which the reservoir input data has been input. An output data generation unit that generates output data by referring to the output result mentioned above. An information processing device equipped with the following features.

[0086] (Aspect B1) A first acquisition unit that acquires training data including input data and label data, A generation unit generates reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition unit that acquires the output result (measurement result) of the quantum reservoir to which the reservoir input data has been input. An output data generation unit that generates output data by referring to the output result, and a learning unit that trains using the label data, An information processing device equipped with the following features.

[0087] (Aspect C1) A quantum reservoir having multiple layers of sub-reservoirs, An acquisition unit that acquires reservoir input data generated from time-series data, A second acquisition unit that acquires the output result (measurement result) of the quantum reservoir after the reservoir input data has been input to the quantum reservoir, An output unit that outputs the aforementioned output result and An information processing device equipped with the following features.

[0088] (Example of an information processing system configuration via the cloud) Below, an example configuration of an information processing system having the above-described structure and using a quantum computer via the cloud will be explained with reference to Figure 6. When this information processing system is applied to robot control, it will be implemented as the robot control system described above.

[0089] As shown in Figure 6, the information processing system comprises a first information processing device 10 and a second information processing device 20. The first information processing device 10 and the second information processing device 20 are configured to communicate with each other via a network N. The first information processing device 10 is, for example, configured to correspond to the determination device 10 described above (excluding configurations corresponding to part or all of the quantum computer 20), and the second information processing device is, for example, configured to correspond to the quantum computer 20 described above or the quantum computing device described above.

[0090] (First information processing device) As shown in Figure 6, the first information processing device 10 includes, as an example, an acquisition unit 12, a generation unit 20a, and an output data generation unit 14.

[0091] The acquisition unit 12 is configured to acquire input data. Here, the input data may be time-series data from sensors installed on the robot, vital data such as brain waves, pulse waves, and respiratory rate of the subject, or sensing data from various sensors installed on the target vehicle, as described above.

[0092] The generation unit 20a generates reservoir input data from the input data acquired by the acquisition unit 12, which is input to a quantum reservoir (reservoir layer 20c in Figure 6) having multiple layers of sub-reservoirs. The generation unit 20a can also be considered as part of the input layer described above. Here, as an example, the generation unit 20a divides the input data into multiple time slices, as described above, and generates the reservoir input data using representative values ​​of the input data in each time slice. Furthermore, as an example, the reservoir input data includes information specifying quantum operations for the qubits constituting the quantum reservoir, as described above.

[0093] The acquisition unit 12 also functions as a second acquisition unit that acquires the output result of the quantum reservoir that has received the reservoir input data.

[0094] The output data generation unit 14 generates output data by referring to the output result of the quantum reservoir. When the acquisition unit 12 is configured to acquire time-series data from sensors provided on the robot as input data, as described above, the output data generation unit 14 generates a control signal corresponding to the output result of the quantum reservoir, as described above, for controlling the operation of the robot.

[0095] Furthermore, if the acquisition unit 12 is configured to acquire vital data such as brain waves, pulse waves, and respiratory rate of the subject as input data, as described above, the output data generation unit 14 generates output data in the form of advice to the subject, such as images or audio, as described above.

[0096] Furthermore, if the acquisition unit 12 is configured to acquire sensing data from various sensors installed on the target vehicle as input data, as described above, the output data generation unit 14 generates control data for controlling the operation of the target vehicle, as described above.

[0097] Furthermore, as described above, the acquisition unit 12 may also acquire label data along with the input data, and the acquisition unit 12 may be configured to function as a learning unit that refers to the label data to train the output data generation unit.

[0098] (Second information processing device) As shown in Figure 6, the second information processing device 20 includes, for example, an acquisition unit 20b, a reservoir layer 20c, and an output unit 20d.

[0099] The acquisition unit 20b acquires the reservoir input data generated by the generation unit 20a of the first information processing device 10. The acquisition unit 20b can also be considered as part of the input layer described above. The reservoir input data acquired by the acquisition unit 20b is input to the reservoir layer 20c.

[0100] The reservoir layer 20c is a quantum reservoir with multiple layers of sub-reservoirs. Since quantum reservoirs have been explained above, their explanation will be omitted here.

[0101] The acquisition unit 20b also functions as a second acquisition unit that acquires the output result of the quantum reservoir after the reservoir input data has been input to the quantum reservoir.

[0102] The output unit 20d outputs the output result. This output result is acquired, for example, by the acquisition unit 12 of the first information processing device 20.

[0103] (Examples of information processing methods in information processing systems) As can be seen from the above description, the information processing according to this embodiment includes, for example, the following information processing methods.

[0104] (Information processing method 1) The acquisition unit 12 performs a first acquisition step in which it acquires input data. The generation unit 20a generates reservoir input data from the input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, in a generation step. The acquisition unit 12 performs a second acquisition step in which it acquires the output result of the quantum reservoir that has received the reservoir input data, and Output data generation step: The output data generation unit 14 generates output data by referring to the output result. An information processing method that includes this.

[0105] (Information processing method 2) The acquisition unit 12 performs a first acquisition step in which it acquires training data including input data and label data. The generation unit 20a generates reservoir input data from the input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, in a generation step. The acquisition unit 12 performs a second acquisition step in which it acquires the output result of the quantum reservoir that has received the reservoir input data, and The acquisition unit 12 (learning unit) trains the output data generation unit, which generates output data by referring to the output results, using the label data in the learning step. An information processing method that includes this.

[0106] (Information processing method 3) The acquisition unit 20b performs a first acquisition step in which it acquires reservoir input data generated from time-series data. A second acquisition step in which the acquisition unit 20b acquires the output result of a quantum reservoir (reservoir layer 20c) having multiple layers of sub-reservoirs after inputting the reservoir input data to the quantum reservoir, and Output step in which the output unit 20d outputs the output result An information processing method that includes this.

[0107] <Additional Note 2> As partially described above, the components of the first information processing device 10 in the above example may be distributed among multiple information processing devices. For example, the first information processing device 10 may be composed of information processing device 10-1 and information processing device 10-2, and information processing device 10-1 is · An acquisition unit 12 as a first acquisition unit for acquiring input data, and • The above-mentioned generation unit 20a The information processing device 10-2 is equipped with, • Acquisition unit 12 as a second acquisition unit for acquiring the output result of the quantum reservoir, • Output data generation unit 14 described above A configuration including the above is also possible. However, the examples of distributed arrangements are not limited to the above examples.

[0108] Similarly, each component of the second information processing device 20 in the above example may be distributed among multiple information processing devices. For example, the second information processing device 20 may be composed of information processing device 20-1 and information processing device 20-2, and information processing device 20-1 may be, • A first acquisition unit 20b for acquiring reservoir input data, and • The reservoir layer 20c described above The information processing device 20-2 is equipped with, · An acquisition unit 20 as a second acquisition unit for acquiring the output result of the reservoir layer 20c, and • Output section 20d described above A configuration including the above is also possible. However, the examples of distributed arrangements are not limited to the above examples.

[0109] Furthermore, the information processing system described above may be configured such that data is supplied from multiple first information processing devices 10 to one or more second information processing devices, and data from the second information processing devices is acquired by the multiple first information processing devices 10. Here, each first information processing device 10 may be configured in a distributed manner as described above. For example, the data from each of the multiple information processing devices 10-1, or data obtained by integrating such data, may be supplied to a second information processing device 20, and data from the second information processing device 20 may be acquired and processed by at least one of the multiple information processing devices 10-2.

[0110] <Additional Note 3> [Examples of implementation using software] The functions of the aforementioned determination devices 10, 10a, quantum computer 20, quantum computing device, and information processing device (hereinafter referred to as "devices") can be realized by programs that cause computers to function as devices, and by programs that cause computers to function as each control block of the devices.

[0111] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., memory) as hardware for executing the program. By executing the program using this control device and storage device, the functions described in each of the embodiments are realized.

[0112] The above program may be recorded on one or more computer-readable recording media, not temporary ones. These recording media may or may not be provided by the above device. In the latter case, the program may be supplied to the above device via any wired or wireless transmission medium.

[0113] Furthermore, some or all of the functions of each of the above control blocks can also be realized by logic circuits. For example, an integrated circuit in which logic circuits that function as each of the above control blocks are formed is also included in the scope of the present invention.

[0114] <Additional Note 4> The inventions described herein include the following configurations:

[0115] (Pattern D1) An acquisition unit that acquires time-series signals from sensors installed on the robot, A determination unit that determines the robot's movement based on the output signal from the sensor, using a reservoir layer which has multiple layers of sub-reservoirs that group multiple qubits, takes the signal as input and processes and learns the robot's movement as output, Based on the determination result, a controller is provided to control the operation of the robot, A robot control system equipped with the following features.

[0116] (Pattern D2) The robot control system according to embodiment D1, wherein the reservoir layer has two or more sub-reservoir layers.

[0117] (Aspect D3) The robot control system according to embodiment D1 or D2, wherein the reservoir layer is a nonlinear physical system that utilizes the dynamics of qubits.

[0118] (Pattern D4) The sensor is a pressure sensor, a vibration sensor, or a camera sensor, according to embodiments D1 to D3. A robot control system as described in one of the following.

[0119] (Pattern D5) A robot comprising a robot control system according to any one of embodiments D1 to D4 and a robot Set system.

[0120] (Pattern D6) The steps include acquiring time-series signals from sensors installed on the robot, The process involves having multiple layers of sub-reservoirs, each containing a group of qubits, and using a reservoir layer that has been trained to process the signal as input and the robot's movement as output, to determine the robot's movement based on the output signal from the sensor. A step of controlling the robot's operation based on the determination result, A robot control method comprising the following features.

[0121] It should be noted that the present invention is not limited to the embodiments described above, and the components can be modified and implemented in practice without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining the multiple components disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiments. Moreover, components from different embodiments may be appropriately combined. [Explanation of symbols]

[0122] 10, 10a Determination device 20 Quantum Computers 30 controllers 40 Robots

Claims

1. A first acquisition unit that acquires input data, A generation unit generates reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition unit that acquires the output result of the quantum reservoir to which the reservoir input data has been input. An output data generation unit that generates output data by referring to the output result mentioned above. An information processing device equipped with the following features.

2. The generating unit is The input data is divided into multiple time slices, and the reservoir input data is generated using representative values ​​of the input data in each time slice. The information processing apparatus according to claim 1.

3. The first acquisition unit further acquires label data along with the input data, The information processing device is A learning unit that trains the output data generation unit by referring to the label data. An information processing apparatus according to claim 1 or 2, further comprising the above.

4. The information processing apparatus according to any one of claims 1 to 3, wherein the quantum reservoir has two or more sub-reservoirs.

5. The information processing apparatus according to any one of claims 1 to 4, wherein the quantum reservoir is a nonlinear physical system that utilizes the dynamics of qubits.

6. The aforementioned input data includes time-series signals from sensors installed on the robot. The quantum reservoir is a trained quantum reservoir that takes the time-series signal as input and outputs the robot's movements. The output data generation unit generates a control signal corresponding to the output result of the quantum reservoir, which is a control signal for controlling the operation of the robot. The information processing apparatus according to any one of claims 1 to 5.

7. The information processing apparatus according to claim 6, wherein the sensor is a pressure sensor, a vibration sensor, or a camera sensor.

8. A robot system comprising the information processing device according to claim 6 or 7 and a robot.

9. A first acquisition unit that acquires training data including input data and label data, A generation unit generates reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition unit that acquires the output result of the quantum reservoir to which the reservoir input data has been input. An output data generation unit that generates output data by referring to the output result, and a learning unit that trains using the label data, An information processing device equipped with the following features.

10. A quantum reservoir having multiple layers of sub-reservoirs, A first acquisition unit that acquires reservoir input data generated from time-series data, A second acquisition unit that acquires the output result of the quantum reservoir after inputting the reservoir input data to the quantum reservoir, An output unit that outputs the aforementioned output result and An information processing device equipped with the following features.

11. An information processing system including a first information processing device and a second information processing device, The first information processing device is A first acquisition unit that acquires input data, A generation unit generates reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition unit that acquires the output result of the quantum reservoir to which the reservoir input data has been input. An output data generation unit that generates output data by referring to the output result mentioned above. Equipped with, The aforementioned second information processing device is A quantum reservoir having multiple layers of sub-reservoirs, A third acquisition unit acquires the reservoir input data generated by the generation unit of the first information processing device, A fourth acquisition unit that acquires the output result of the quantum reservoir after inputting the reservoir input data to the quantum reservoir, An output unit that outputs the aforementioned output result and An information processing system equipped with the following features.

12. One or more processors, The first acquisition step involves obtaining input data, A generation step of generating reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition step involves obtaining the output result of the quantum reservoir to which the reservoir input data has been input. Output data generation step: Generates output data by referring to the output result mentioned above. An information processing method that includes this.

13. One or more processors, A first acquisition step involves obtaining training data that includes input data and label data, A generation step of generating reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition step involves obtaining the output result of the quantum reservoir to which the reservoir input data has been input. A learning step in which an output data generation unit that generates output data by referring to the output result is trained using the label data, An information processing method that includes this.

14. One or more processors, The first acquisition step involves obtaining reservoir input data generated from time-series data, A second acquisition step involves obtaining the output result of a quantum reservoir after inputting the reservoir input data to a quantum reservoir having multiple layers of sub-reservoirs, Output step to output the aforementioned output result An information processing method that includes this.

15. A program for causing a computer to function as an information processing device, wherein the computer, The first acquisition step involves obtaining input data, A generation step of generating reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition step involves obtaining the output result of the quantum reservoir to which the reservoir input data has been input. Output data generation step: Generates output data by referring to the output result mentioned above. A program that executes the command.

16. A program for causing a computer to function as an information processing device, wherein the computer, A first acquisition step involves obtaining training data that includes input data and label data, A generation step of generating reservoir input data from the aforementioned input data to be input to a quantum reservoir having multiple layers of sub-reservoirs, A second acquisition step involves obtaining the output result of the quantum reservoir to which the reservoir input data has been input. A learning step in which an output data generation unit that generates output data by referring to the output result is trained using the label data, A program that executes the command.

Images