Data processing method, information processing device, and program

Abstract

To improve a technique related to data processing using a neural network model.SOLUTION: A data processing method executed by an information processing device includes the step of: causing a computer to train a first neural network model, a second neural network model, and a third neural network model using first data, second data, and third data as training data such that the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model satisfy a first constraint condition. The first constraint condition is a condition that trains the computer to express the third data on the basis of the first and second weight coefficients.SELECTED DRAWING: Figure 4

Description

Technical Field

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

Background Art

[0002] Conventionally, data processing and conversion techniques using neural network models such as Implicit Neural Representations (INR) have been known (for example, Non-Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Even when a neural network model learns the same matter, there are extremely high redundancies, such as various combinations of weight coefficients obtained as a result of learning depending on the learning process, the setting of the initial values of the weight coefficients, etc. However, the conventional INR has not considered at all utilizing such redundancies. On the other hand, by utilizing redundancies, for example, it is conceivable to give a certain relationship between the weight coefficients of a plurality of learning models. Thus, there has been room for improvement in the technology related to data processing and conversion using neural network models.

[0005] In view of such circumstances, an object of the present disclosure is to improve the technology related to data processing and conversion using neural network models.

Means for Solving the Problems

[0006] (1) A data representation method according to an embodiment of the present disclosure is a data processing method executed by an information processing apparatus, comprising: using first data, second data, and third data as teacher data, and training a first neural network model, a second neural network model, and a third neural network model, respectively, such that a first weight coefficient related to the first neural network model, a second weight coefficient related to the second neural network model, and a third weight coefficient related to the third neural network model satisfy a first constraint condition; including: wherein the first constraint condition is a condition for training to represent the third data based on the first weight coefficient and the second weight coefficient.

[0007] (2) A data processing method according to an embodiment of the present disclosure is the data processing method according to (1), wherein the third weight coefficient is an average value of the first weight coefficient and the second weight coefficient.

[0008] (3) A data processing method according to an embodiment of the present disclosure is the data processing method according to (1), wherein the third weight coefficient is a weighted average value of the first weight coefficient and the second weight coefficient.

[0009] (4) A data processing method according to an embodiment of the present disclosure is the data processing method according to any one of (1) to (3), further comprising: using fourth data as teacher data, and training a fourth neural network model such that the first weight coefficient, the second weight coefficient, and a fourth weight coefficient related to the fourth neural network model satisfy a second constraint condition; including: wherein the second constraint condition is a condition for training to represent the fourth data based on the first weight coefficient and the second weight coefficient.

[0010] ​(5) The data processing method according to an embodiment of the present disclosure is the data processing method according to any one of (1) to (3), and The activation functions of the first neural network model, the second neural network model, and the third neural network model are Sine functions.

[0011] (6) The data processing method according to an embodiment of the present disclosure is the data representation method according to any one of (1) to (5), and In the step of learning, the initial value of the first weight coefficient and the initial value of the second weight coefficient are the same.

[0012] (7) The data processing method according to an embodiment of the present disclosure is the data processing method according to any one of (1) to (6), and The number of layers of the first neural network model, the number of layers of the second neural network model, and the number of layers of the third neural network model are the same.

[0013] (8) The data processing method according to an embodiment of the present disclosure is A data processing method executed by an information processing apparatus, Using first data, second data, and third data as teacher data, respectively, learning the first neural network model, the second neural network model, and the third neural network model so that the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model satisfy a first constraint condition, where the first constraint condition is a condition for learning to represent the third data based on the first weight coefficient and the second weight coefficient, and obtaining the first weight coefficient and the second weight coefficient; Generating the third weight coefficient based on the first weight coefficient and the second weight coefficient; Performing data processing based on the third weight coefficient; including.

[0014] (9) An information processing apparatus according to an embodiment of the present disclosure is an information processing apparatus including a control unit, wherein the control unit uses first data, second data, and third data as teacher data, and respectively trains a first neural network model, a second neural network model, and a third neural network model so that a first weight coefficient related to the first neural network model, a second weight coefficient related to the second neural network model, and a third weight coefficient related to the third neural network model satisfy a first constraint condition, wherein the first constraint condition is a condition for training to represent the third data based on the first weight coefficient and the second weight coefficient.

[0015] (10) A program according to an embodiment of the present disclosure causes a computer to use first data, second data, and third data as teacher data, and respectively train a first neural network model, a second neural network model, and a third neural network model so that a first weight coefficient related to the first neural network model, a second weight coefficient related to the second neural network model, and a third weight coefficient related to the third neural network model satisfy a first constraint condition, wherein the first constraint condition is a condition for training to represent the third data based on the first weight coefficient and the second weight coefficient.

Advantages of the Invention

[0016] According to an embodiment of the present disclosure, the technology related to data processing and conversion using a neural network model is improved.

Brief Description of the Drawings

[0017]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Mode for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present disclosure will be described.

[0019] (Outline of Embodiment) With reference to FIG. 1, the outline and configuration of the system 1 according to the present embodiment will be described.

[0020] The system 1 according to this embodiment includes an information processing device 10a and an information processing device 10b. The information processing device 10a and the information processing device 10b are communicably connected to a network 20 including, for example, a mobile communication network and the Internet. The information processing device 10a and the information processing device 10b do not necessarily need to be constantly communicably connected, and data may be transmitted, for example, by an information storage medium such as a memory. In this specification, when the information processing device 10a and the information processing device 10b are not distinguished, they are collectively referred to as the information processing device 10.

[0021] The information processing device 10 is an arbitrary device used by a user. A general-purpose computer or a dedicated computer can be adopted as the information processing device 10. Although FIG. 1 shows an example in which the system 1 includes one information processing device 10, the present invention is not limited to this. The system 1 may include two or more information processing devices 10.

[0022] First, the outline of the present technology will be described in the case where the information processing device 10a and the information processing device 10b each function as an encoding device and a decoding device. For example, the information processing device 10a encodes data such as an image, and the information processing device 10b decodes data such as text, an image, and audio from the encoded signal. Here, in this embodiment, the image includes a still image and a moving image.

[0023] In the encoding process and the decoding process according to this embodiment, a data representation technique using a neural network model is used. For example, the information processing device 10a uses image data as teacher data to train a neural network model 30a. In FIG. 1, as an example, an image with a size of 256×256 is used as teacher data. The coordinates (x, y) of each pixel are normalized to a value (p in the range of, for example, [-1, 1]. x p y) is input into the neural network model 30a. The weight coefficient θ1 of the neural network model 30a (θ1 includes the weights themselves and the bias b1 and refers to all the parameters of the neural network; the same applies hereinafter) is updated based on the learning process so as to output the pixel values (RGB) of the image which is the teacher data.

[0024] The information processing device 10b decodes the image data using the weight coefficient θ1. In other words, the image data is represented by the weight coefficient θ1. The data of the weight coefficient θ1 is transmitted from the information processing device 10a to the information processing device 10b via the network 20. In FIG. 1, as an example, an example of decoding after enlarging to an image of size 1024×1024 is shown. The coordinates (x, y) of each pixel are normalized to values (p x , p y ) that are input into the neural network model 30b. By using the weight coefficient θ1, the image data of the teacher data used for the learning of the information processing device 10a is enlarged and decoded. Here, an example of enlargement and decoding by the information processing device 10b has been described, but it is not limited to this. The information processing device 10b may decode an image of the same size (256×256 size) as the original image data, or may also reduce it.

[0025] The system 1 according to this embodiment relates to the above-described technology and improves the data representation technology using a neural network model. Briefly, the information processing device 10 uses the first data, the second data, and the third data as teacher data, and respectively uses the first neural network model, the second neural network model, and the third neural network model to satisfy the first constraint condition for the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model. to learn so that Hereinafter, in the present embodiment, the entire learning model including the first neural network model, the second neural network model, and the third neural network model is also referred to as a hybrid model. Here, the first constraint condition is a condition for learning to represent the third data based on the first weight coefficient and the second weight coefficient. In other words, the information processing device 10 according to the present embodiment represents the third data based on the first weight coefficient and the second weight coefficient.

[0026] Thus, according to the present embodiment, the first neural network model, the second neural network model, and the third neural network model are learned so that the first weight coefficient, the second weight coefficient, and the third weight coefficient satisfy the first constraint condition. That is, in the present embodiment, the redundancy of the neural network model is utilized to give a predetermined relationship to the first weight coefficient, the second weight coefficient, and the third weight coefficient. As a result, the third data is represented based on the first weight coefficient and the second weight coefficient. Therefore, for example, if only the first weight coefficient and the second weight coefficient are transmitted to the information processing device on the decoding side, the third data can be decoded, which is advantageous from the viewpoint of improving the efficiency of data transmission. In other words, the technology related to data representation using a neural network model is improved in that the third data can be decoded by transmitting the weight coefficient of the first data and the weight coefficient of the second data.

[0027] (Configuration of Information Processing Device) As shown in FIG. 2, the information processing apparatus 10 includes a control unit 11, a storage unit 12, an input unit 13, an output unit 14, and a communication unit 15.

[0028] The control unit 11 includes at least one processor, at least one dedicated circuit, or a combination thereof. The processor is a general-purpose processor such as a CPU (central processing unit) or a GPU (graphics processing unit), or a dedicated processor specialized for specific processing. The dedicated circuit is, for example, an FPGA (field-programmable gate array) or an ASIC (application specific integrated circuit). The control unit 11 executes processes related to the operation of the information processing apparatus 10 while controlling each unit of the information processing apparatus 10.

[0029] The storage unit 12 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or a combination of at least two of these. The semiconductor memory is, for example, a RAM (random access memory) or a ROM (read only memory). The RAM is, for example, an SRAM (static random access memory) or a DRAM (dynamic random access memory). The ROM is, for example, an EEPROM (electrically erasable programmable read only memory). The storage unit 12 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 12 stores data used for the operation of the information processing apparatus 10 and data obtained by the operation of the information processing apparatus 10.

[0030] The input unit 13 includes at least one input interface. The input interface is, for example, a physical key, a capacitive key, a pointing device, or a touch screen provided integrally with a display. The input interface may also be, for example, a sound sensor that receives voice input, or a camera that receives gesture input. The input unit 13 receives an operation for inputting data used in the operation of the information processing apparatus 10. Instead of being provided in the information processing apparatus 10, the input unit 13 may be connected to the information processing apparatus 10 as an external input device. As a connection method, for example, any method such as USB (Universal Serial Bus), HDMI (Registered Trademark) (High-Definition Multimedia Interface), or Bluetooth (Registered Trademark) can be used.

[0031] The output unit 14 includes at least one output interface. The output interface is, for example, a display that outputs information as video, or a speaker that outputs information as sound. The display is, for example, an LCD (liquid crystal display) or an organic EL (electro luminescence) display. The output unit 14 performs display output of data obtained by the operation of the information processing apparatus 10. Instead of being provided in the information processing apparatus 10, the output unit 14 may be connected to the information processing apparatus 10 as an external output device. As a connection method, for example, any method such as USB, HDMI (Registered Trademark), or Bluetooth (Registered Trademark) can be used.

[0032] The communication unit 15 includes at least one interface for external communication. The communication interface may be either a wired communication interface or a wireless communication interface. In the case of wired communication, the communication interface is, for example, a LAN (Local Area Network) interface or a USB (Universal Serial Bus). In the case of wireless communication, the communication interface is, for example, an interface corresponding to a mobile communication standard such as LTE (Long Term Evolution), 4G (4th generation), or 5G (5th generation), or an interface corresponding to short-range wireless communication such as Bluetooth (registered trademark). The communication unit 15 receives data used for the operation of the information processing apparatus 10 and transmits data obtained by the operation of the information processing apparatus 10.

[0033] The functions of the information processing apparatus 10 are realized by causing a processor corresponding to the information processing apparatus 10 to execute the program according to this embodiment. That is, the functions of the information processing apparatus 10 are realized by software. The program causes a computer to execute the operations of the information processing apparatus 10, thereby causing the computer to function as the information processing apparatus 10. That is, the computer functions as the information processing apparatus 10 by executing the operations of the information processing apparatus 10 according to the program.

[0034] In this embodiment, the program can be recorded on a computer-readable recording medium. The computer-readable recording medium includes a non-transitory computer-readable medium, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, or a semiconductor memory. The distribution of the program is performed, for example, by selling, transferring, or lending a portable recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory) on which the program is recorded. The distribution of the program may also be performed by storing the program in the storage of an external server and transmitting the program from the external server to another computer. The program may also be provided as a program product.

[0035] Some or all of the functions of the information processing apparatus 10 may be realized by a dedicated circuit corresponding to the control unit 11. That is, some or all of the functions of the information processing apparatus 10 may be realized by hardware.

[0036] (Operation of the information processing apparatus) With reference to FIGS. 3 and 7, the operation of the information processing apparatus 10 according to this embodiment will be described. First, with reference to FIG. 3, the step of encoding an image in the information processing apparatus 10a will be described.

[0037] Step S100: The control unit 11 of the information processing apparatus 10a uses the first data, the second data, and the third data as teacher data to train a first neural network model, a second neural network model, and a third neural network model, respectively, such that the first weight coefficient, the second weight coefficient, and the third weight coefficient satisfy the first constraint condition. That is, the control unit 11 trains the first neural network model, the second neural network model, and the third neural network model with constraints. The first constraint condition is a condition for training to represent the third data 131 based on the first weight coefficient and the second weight coefficient. Here, the first weight coefficient, the second weight coefficient, and the third weight coefficient include the weights themselves and biases, and refer to all the parameters of the neural network.

[0038] The first data, the second data, and the third data may all be arbitrary data such as text, images, and sounds. Hereinafter, in the present embodiment, it is described that the first data, the second data, and the third data are all image data, but the present invention is not limited thereto.

[0039] The first constraint condition may be, for example, to train such that the third weight coefficient is the average value of the first weight coefficient and the second weight coefficient. Alternatively, the first constraint condition may be to train such that the third weight coefficient is the weighted average value of the first weight coefficient and the second weight coefficient. Or, the first constraint condition may be to train such that the third weight coefficient is a value obtained by applying a function to the first and second weight coefficients. The first constraint condition may be any condition as long as it trains the third weight coefficient to be represented by the first weight coefficient and the second weight coefficient.

[0040] Fig. 4 shows a schematic configuration of the hybrid model 100 according to the present embodiment. As shown in Fig. 4, the hybrid model 100 according to the present embodiment includes a first neural network model 110, a second neural network model 120, and a third neural network model 130. The first neural network model 110, the second neural network model 120, and the third neural network model 130 are each learned with first data 111, second data 121, and third data 131 as teacher data so that the respective weight coefficients θ1, θ2, and θ3 satisfy a first constraint condition. In the present embodiment, that the respective weight coefficients θ1, θ2, and θ3 satisfy the first constraint condition corresponds to that θ3 is determined by a function of θ1 and θ2. For example, when the third weight coefficient is the average value of the first weight coefficient and the second weight coefficient, θ3 is represented by the following mathematical formula (1).

[0041]

Number

[0042] Also, for example, when the third weight coefficient is the weighted average value of the first weight coefficient and the second weight coefficient, θ3 is represented by the following mathematical formula (2).

[0043]

Number

[0044] Specifically, the learning process inputs the coordinates (x, y) of each pixel into the neural network model, and repeats the learning a predetermined number of times (number of epochs) to repeatedly update the weight coefficients θ1, θ2, and θ3 so that the loss function is minimized. A conceptual diagram of the learning process is shown in FIG. 5. Based on the error between the output obtained by inputting the coordinates (x, y) of each pixel into the first neural network model 110 and the first data 111 which is the teacher data, the value of the loss function corresponding to the first neural network model 110 is determined. Based on the error between the output obtained by inputting the coordinates (x, y) of each pixel into the second neural network model 120 and the second data 121 which is the teacher data, the value of the loss function corresponding to the second neural network model 120 is determined. Also, based on the error between the output obtained by inputting the coordinates (x, y) of each pixel into the third neural network model 130 and the third data 131 which is the teacher data, the value of the loss function corresponding to the third neural network model 130 is determined. The weight coefficients θ1, θ2, and θ3 are updated so that the value calculated from the values of these three loss functions becomes the minimum. The value calculated from the values of the three loss functions is, for example, the average value of the values of the three loss functions, but is not limited to this. Here, the initial values of the first weight coefficient and the second weight coefficient (the initial values of θ1 and θ2) are determined randomly. Although the initial value of the first weight coefficient and the initial value of the second weight coefficient are often the same, this is not the only case. On the other hand, the initial value of the third weight coefficient (the initial value of θ3) is determined so as to satisfy the first constraint condition. In such a learning process, the update of the weight coefficients θ1 and θ2 is executed based on the error backpropagation method (backpropagation). The weight coefficient θ3 is updated so as to satisfy the first constraint condition.

[0045] As a result of the learning process, the first weight coefficient θ1 and the second weight coefficient θ2 are updated, and the third weight coefficient is determined based on the updated first weight coefficient θ1 and second weight coefficient θ2. When the coordinates (x, y) of each pixel are input to each neural network model constituting such a hybrid model 100, data corresponding to the first data 111, the second data 121, and the third data 131 are output from the first neural network model 110, the second neural network model 120, and the third neural network model 130, respectively. Note that the data corresponding to the first data 111, the second data 121, and the third data 131 are data that reproduce the first data 111, the second data 121, and the third data 131, which are teacher data. In other words, the data corresponding to the first data 111, the second data 121, and the third data 131 are data that are the same as or similar to the first data 111, the second data 121, and the third data 131, which are teacher data.

[0046] FIG. 6 is a schematic diagram showing a hybrid model in which the third weight coefficient is the average value of the first weight coefficient and the second weight coefficient as the first constraint condition. In FIG. 6, the first constraint condition is represented as the function F α and is shown. In the hybrid model shown in FIG. 6, the third weight coefficient θ3 is represented by the following equation (3).

Equation

[0047] As described above, the information processing apparatus 10a can encode the first data, the second data, and the third data with the first weight coefficient and the second weight coefficient.

[0048] Returning to FIG. 3, the operation of the information processing apparatus 10a will be described. Step S110: The output unit 14 of the information processing 10a outputs the first weight coefficient and the second weight coefficient generated in step S100.

[0049] Next, with reference to FIG. 7, the operation of the information processing apparatus 10b will be described. The information processing apparatus 10b decodes the data encoded by the information processing apparatus 10a. Step S200: The input unit 13 of the information processing apparatus 10b acquires the first weight coefficient and the second weight coefficient generated by the information processing apparatus 10a.

[0050] Step S210: Next, the control unit 11 generates a third weight coefficient based on the first weight coefficient and the second weight coefficient. The third weight coefficient generated by the information processing apparatus 10b is generated in the same manner as the generation of the third weight coefficient by the information processing apparatus 10a. For example, when the information processing apparatus 10a calculates the third weight coefficient by averaging the first weight coefficient and the second weight coefficient, the information processing apparatus 10b also calculates the third weight coefficient by averaging the first weight coefficient and the second weight coefficient. Step S220: Next, the control unit 11 sets the first weight coefficient, the second weight coefficient, and the third weight coefficient in the neural network model respectively, inputs the coordinates (x, y) of each pixel, and the obtained output becomes data corresponding to the first data, the second data, and the third data respectively, and decoding is performed.

[0051] As described above, the information processing apparatus 10 according to the present embodiment causes the first neural network model, the second neural network model, and the third neural network model to be learned so that the first weight coefficient, the second weight coefficient, and the third weight coefficient satisfy the constraints. According to such a configuration, the third data is expressed based on the first weight coefficient and the second weight coefficient. Therefore, for example, if only the first weight coefficient and the second weight coefficient are transmitted to the information processing apparatus on the decoding side, the third data can be decoded, which has an advantage from the viewpoint of improving the efficiency of data transmission and the like. In other words, the technology related to data representation using the neural network model is improved in that the third data can be decoded by transmitting the weight coefficient of the first data and the weight coefficient of the second data.

[0052] In the present embodiment, the case where the hybrid model includes three neural network models has been shown. However, the number of neural network models included in the hybrid model may be four or more. FIG. 7 is a schematic diagram showing an example of the configuration when the hybrid model includes four neural network models. The hybrid model 100B shown in FIG. 7 includes a first neural network model 110, a second neural network model 120, a third neural network model 130, and a fourth neural network model 140. The first neural network model 110, the second neural network model 120, the third neural network model 130, and the fourth neural network model 140 are respectively learned by the first data 111, the second data 121, the third data 131, and the fourth data 141 as teacher data so as to satisfy the weight coefficients θ1, θ2, and θ B are learned to satisfy the first constraint condition. As described above, the first constraint condition is a condition for learning to represent the third data 131 based on the first weight coefficient and the second weight coefficient. Also, the weight coefficients θ1, θ2, and θ C are learned to satisfy the second constraint condition. The second constraint condition is a condition for learning to represent the fourth data 141 based on the first weight coefficient and the second weight coefficient. In FIG. 7, the first constraint condition is represented as the function F β . That each of the weight coefficients θ1, θ2, and θ B satisfies the first constraint condition means that θ B is determined by a function of θ1 and θ2. The weight coefficient θ B of the hybrid model 100B shown in FIG. 7 is represented by the following mathematical formula (4).

Equation

[0053] Also, in FIG. 7, the first constraint condition is represented as the function F γ . That each of the weight coefficients θ1, θ2, and θ C satisfies the second constraint condition means that θ C is determined by a function of θ1 and θ2. The weight coefficient θ of the hybrid model 100B shown in FIG. 7C It is represented by the following mathematical formula (5).

Number

[0054] Thus, in this embodiment, the first neural network model, the second neural network model, the third neural network model, and the fourth neural network model may be learned so that each weight coefficient satisfies the constraint. According to such a configuration, the third data and the fourth data are represented based on the first weight coefficient and the second weight coefficient. Therefore, for example, if only the first weight coefficient and the second weight coefficient are transmitted to the information processing device on the decoding side, the third data and the fourth data can be decoded, which has advantages from the viewpoint of improving the efficiency of data transmission and the like. There are advantages such as this. In other words, the technique regarding data representation using the neural network model is improved in that the third data and the fourth data can be decoded by transmitting the weight coefficients of the first data and the second data.

[0055] FIG. 8 is a schematic diagram showing a case where each teacher data is a partial image obtained by dividing the entire image 200 into four parts in the hybrid model 100B. For example, in such a case, the four partial images (partial image 211, partial image 221, partial image 231, and partial image 241) are represented based on the first weight coefficient and the second weight coefficient. Therefore, for example, if only the first weight coefficient and the second weight coefficient are transmitted to the information processing device on the decoding side, all the partial images can be decoded, and as a result, there are advantages such as the entire image 200 can be decoded. Note that even when the image is divided into other than four parts, the same processing can be performed as described above.

[0056] In this embodiment, an example in which the number of neural network models included in the hybrid model is 3 or 4 has been described, but the present invention is not limited to this. The number of neural network models included in the hybrid model may be 5 or more. By using the technique of the present disclosure to give the relationship of the weight coefficients, generally, different weight coefficients θ1,..., θ nBased on this, m pieces of data (where m > n) can be represented. Here, the number of layers of each neural network model included in the hybrid model may or may not be the same. When the number of layers of each neural network model is the same, there is an advantage in that the constraint conditions between the weight coefficients are simplified. For example, when the hybrid model includes three neural network models, the number of layers of the first neural network model, the number of layers of the second neural network model, and the number of layers of the third neural network model may be the same. By doing so, the constraint conditions for the weight coefficients of each neural network model can be easily set. On the other hand, even when the number of layers of each neural network model is not the same, a constraint equation can be defined such that θ3 is obtained from the calculation of θ1 and θ3.

[0057] Note that when encoding image data, the activation function of each neural network model included in the hybrid model is preferably a Sine function. By adopting the Sine function as the activation function, the teacher data can be accurately reproduced. Note that the activation function may be appropriately changed according to the object to be encoded. For example, the activation function may be any function such as a ReLU function or a sigmoid function.

[0058] Although the present disclosure has been described based on the drawings and examples, it should be noted that those skilled in the art may make various modifications and alterations based on the present disclosure. Therefore, it should be noted that these modifications and alterations are included in the scope of the present disclosure. For example, the functions included in each component or each step, etc. can be rearranged so as not to be logically contradictory, and a plurality of components or steps, etc. can be combined into one or divided. Also, although the present disclosure has been described based on still images, not only image data but also data that can be encoded by a neural network model can be encoded and decoded based on the present disclosure. Further, it is conceivable to apply the technology of the present disclosure to image classification and the like using a general neural network. For example, it is also conceivable to improve the adversarial attack resistance by using a third neural network corresponding to the first neural network and the second neural network.

Industrial Applicability

[0059] The technology of the present disclosure can be used not only for data representation technologies such as images and data encoding technologies but also for security technologies and the like.

Explanation of Signs

[0060] 1 System 10, 10a, 10b Information Processing Apparatus 11 Control Unit 12 Storage Unit 13 Input Unit 14 Output Unit 15 Communication Unit 20 Network 30a, 30b Neural Network Model 100, 100B Hybrid Model 110 First Neural Network Model 120 Second Neural Network Model 130 Third Neural Network Model 140 Fourth Neural Network Model 111 First Data 121 Second Data 131 Third Data 141 Fourth Data 200 Whole Image 211, 221, 231, 241 Partial Image Data

Claims

1. A data processing method executed by an information processing apparatus, comprising: using first data, second data, and third data as teacher data, and training a first neural network model, a second neural network model, and a third neural network model respectively so that a first weight coefficient related to the first neural network model, a second weight coefficient related to the second neural network model, and a third weight coefficient related to the third neural network model satisfy a first constraint condition; including wherein the first constraint condition is a condition for training to represent the third data based on the first weight coefficient and the second weight coefficient.

2. The data processing method according to claim 1, wherein the third weight coefficient is an average value of the first weight coefficient and the second weight coefficient.

3. The data processing method according to claim 1, wherein the third weight coefficient is a weighted average value of the first weight coefficient and the second weight coefficient.

4. The data processing method according to claim 1, further comprising: using fourth data as teacher data and training a fourth neural network model so that the first weight coefficient, the second weight coefficient, and a fourth weight coefficient related to the fourth neural network model satisfy a second constraint condition; wherein the second constraint condition is a condition for training to represent the fourth data based on the first weight coefficient and the second weight coefficient.

5. The data processing method according to claim 1, wherein activation functions of the first neural network model, the second neural network model, and the third neural network model are Sine functions.

6. The data processing method according to claim 1, wherein in the training step, an initial value of the first weight coefficient and an initial value of the second weight coefficient are the same.

7. The data processing method according to claim 1, wherein the number of layers of the first neural network model, the number of layers of the second neural network model, and the number of layers of the third neural network model are the same.

8. A data processing method executed by an information processing apparatus, comprising: Using the first data, the second data, and the third data as teacher data, respectively, the first neural network model, the second neural network model, and the third neural network model are learned so that the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model satisfy the first constraint condition. The first constraint condition is a condition for learning to represent the third data based on the first weight coefficient and the second weight coefficient. A step of obtaining the first weight coefficient and the second weight coefficient; A step of generating the third weight coefficient based on the first weight coefficient and the second weight coefficient; A step of performing data processing based on the third weight coefficient; A data processing method including the above.

9. An information processing apparatus including a control unit, The control unit, Using the first data, the second data, and the third data as teacher data, respectively, the first neural network model, the second neural network model, and the third neural network model are learned so that the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model satisfy the first constraint condition. The first constraint condition is a condition for learning to represent the third data based on the first weight coefficient and the second weight coefficient. An information processing apparatus.

10. A program for causing a computer to Using the first data, the second data, and the third data as teacher data, respectively, the first neural network model, the second neural network model, and the third neural network model are learned so that the first weight coefficient related to the first neural network model, the second weight coefficient related to the second neural network model, and the third weight coefficient related to the third neural network model satisfy the first constraint condition. Execute, The first constraint condition is to learn to represent the third data based on the first weight coefficient and the second weight coefficient. Program.

Images