JP7350214B2 - Inference device, inference method, and inference program - Google Patents

Description

The present disclosure relates to an inference device, an inference method, and an inference program.

When installing a deep learning inference environment on a device with resource constraints such as an embedded device, it is necessary to reduce the weight of deep learning calculations. One method for reducing the weight is to replace floating point operations with fixed point operations or integer operations. This technique is commonly called quantization. Patent Document 1 discloses a technique of estimating data distribution using a statistical method in order to perform quantization with relatively high accuracy to speed up inference.

Japanese Patent Application Publication No. 2018-010618

According to the technology disclosed in Patent Document 1, in quantization, parameters of calculations related to inference are adjusted so that overflow does not occur when a learning data set or the like is used. However, the training data set etc. cannot cover all inference data. Therefore, according to the technology disclosed in Patent Document 1, it cannot be guaranteed that overflow will not occur no matter what kind of inference data is used, and overflow may occur depending on the inference data. There is a problem. Here, since the inference data may be confidential information, even if an overflow occurs when inference is performed using a certain inference data, the overflow that has occurred can be handled using the inference data. It is not always possible to analyze the

The present disclosure provides a method for acquiring data for analyzing the overflow that is different from the certain inference data when an overflow occurs when inference is performed using certain inference data. purpose.

The inference device according to the present disclosure includes:
a quantization inference unit that performs at least one quantization operation based on a machine learning method using the inference data;
An inference device comprising: a non-quantization inference unit that executes at least one of at least one non-quantization operation corresponding to each of the at least one quantization operation using the inference data,
Each of the at least one quantization operation is an operation corresponding to at least one quantization feature data indicating a characteristic of each of the at least one quantization operation,
Each of the at least one non-quantized operation is an operation according to each of at least one non-quantized feature data indicating the characteristics of each of the at least one non-quantized operation,
The inference device further includes:
When an overflow occurs in at least one of the at least one quantization operation, quantization feature data corresponding to each quantization operation in which an overflow occurred, and quantization feature data corresponding to each quantization operation in which an overflow occurred. The apparatus includes a data extraction unit that extracts non-quantized feature data corresponding to non-quantized operations.

According to the present disclosure, when an overflow occurs in inference based on a machine learning method , the data extraction unit extracts quantized feature data and non-quantized feature data related to the overflow that has occurred. . Here, each of the quantized feature data and the non-quantized feature data is data different from the inference data. Therefore, according to the present disclosure, when an overflow occurs when inference is performed using certain inference data, data for analyzing the overflow that has occurred and which is different from the certain inference data is used. can be obtained.

1 is a diagram showing a configuration example of an inference device 100 according to Embodiment 1. FIG. FIG. 3 is a diagram illustrating priority levels of processes according to the first embodiment. 1 is a diagram showing an example of a hardware configuration of an inference device 100 according to Embodiment 1. FIG. 7 is a flowchart showing the operation of the inference device 100 according to the first embodiment. FIG. 3 is a diagram illustrating the operation of the inference device 100 according to the first embodiment. 7 is a diagram illustrating an example of the hardware configuration of an inference device 100 according to a modification of the first embodiment. FIG. 7 is a diagram illustrating an example of the hardware configuration of an inference device 100 according to a modification of the first embodiment. FIG. FIG. 3 is a diagram illustrating a configuration example of an inference device 100 according to a second embodiment. 7 is a flowchart showing the operation of the inference device 100 according to the second embodiment.

In the description of the embodiments and the drawings, the same elements and corresponding elements are denoted by the same reference numerals. Descriptions of elements labeled with the same reference numerals will be omitted or simplified as appropriate. Arrows in the figure mainly indicate the flow of data or processing. Furthermore, "unit" may be read as "circuit," "process," "procedure," "process," or "circuitry" as appropriate.

Embodiment 1.
Hereinafter, this embodiment will be described in detail with reference to the drawings.

***Explanation of configuration***
FIG. 1 shows a configuration example of an inference device 100 according to this embodiment. As shown in this figure, the inference device 100 includes a quantized inference section 110, a non-quantized inference section 120, and a data extraction section 130.
Reasoning device 100 is typically part of an embedded system. Note that the inference device 100 may include an inference process management section, and the inference process management section may control the quantization inference section 110 and the non-quantization inference section 120.

The quantization inference unit 110 executes a quantization inference process, that is, executes at least one quantization operation based on a machine learning technique. Each of the at least one quantization operation is an operation corresponding to at least one piece of quantized feature data. Each of the at least one quantization feature data indicates a feature of each of at least one quantization operation, and may also include a parameter corresponding to each of at least one quantization operation. Further, the quantization inference unit 110 uses inference data. Inference data is data that is input to the learned model when performing inference.
The quantized inference process is also called the quantized inference process. A quantization algorithm is executed in the quantization inference process. Quantization algorithms are also called quantization inference algorithms. Furthermore, when an overflow or the like occurs in the quantization inference process, the quantization inference unit 110 records information regarding the overflow or the like that has occurred.

The non-quantized inference unit 120 executes a non-quantized inference process, that is, executes at least one non-quantized operation corresponding to each of the at least one quantized operation. At least one quantization operation and at least one non-quantization operation are operations based on the same machine learning method. When the machine learning method is a neural network, each quantization operation and each non-quantization operation are operations for each layer of the neural network. Each of the at least one non-quantized operation is an operation corresponding to at least one non-quantized feature data. Each of the at least one non-quantized feature data indicates a feature of each of the at least one non-quantized operation, and may also include a parameter corresponding to each of the at least one non-quantized operation. The non-quantized inference unit 120 uses inference data and saved data, which will be described later.
A non-quantized inference process is also called an unquantized inference process. A non-quantized algorithm is executed in the non-quantized inference process. Non-quantized algorithms are also called non-quantized inference algorithms. The non-quantized algorithm is an algorithm obtained as a result of learning using learning data. The quantization algorithm and the non-quantization algorithm are basically the same. The quantization algorithm is a non-quantization algorithm modified as appropriate to support quantization. Quantization typically involves replacing floating point operations with fixed point or integer operations. Further, each of the quantization algorithm and the non-quantization algorithm may be generated by the inference device 100, or may be generated by another device. Further, the non-quantization inference unit 120 executes calculations according to the input data used when an overflow occurs in the quantization inference process.

The data extraction unit 130 extracts saved data from the quantized inference process and the non-quantized inference process when an overflow occurs in the quantized inference process. The saved data is data used when analyzing overflow in the quantization inference process, and includes input data and feature data as a specific example. The input data is data that is input when performing at least one quantization operation and at least one non-quantization operation. Feature data is data utilized in quantization calculations. Feature data is a general term for quantized feature data and non-quantized feature data, and exists for each operation: at least one quantization operation and at least one non-quantization operation. represents the amplitude of input data in . The amplitude of the input data is a range from the minimum value to the maximum value of the values assumed as the input data. When an overflow occurs in at least one of at least one quantization operation, the data extraction unit 130 extracts quantization feature data corresponding to each of the quantization operations in which the overflow occurred and the quantization operation in which the overflow occurred. The non-quantized feature data corresponding to the non-quantized operation corresponding to each of the non-quantized feature data are extracted.
When the machine learning method is deep learning , the data extraction unit 130 may extract data indicating parameters for a layer corresponding to a quantization operation in which an overflow has occurred as quantization feature data; Data indicating parameters for a layer corresponding to a non-quantized operation corresponding to a quantized operation in which an overflow has occurred may be extracted as the quantized feature data.

FIG. 2 shows the priority of each process executed by the inference device 100. As shown in the figure, the priority of the quantized inference process is higher than the priority of the non-quantized inference process. Furthermore, there may be a process that has a lower priority than the quantized inference process and a higher priority than the non-quantized inference process. Note that the parentheses surrounding the notation ``other process'' indicate that the other process may or may not be present.

FIG. 3 shows an example of the hardware configuration of the inference device 100 according to this embodiment. Reasoning device 100 consists of a computer. The inference device 100 may consist of multiple computers.

As shown in the figure, the inference device 100 is a computer that includes hardware such as a processor 11, a memory 12, an auxiliary storage device 13, an input/output IF (Interface) 14, and a communication device 15. These pieces of hardware are appropriately connected via signal lines 19.

The processor 11 is an IC (Integrated Circuit) that performs arithmetic processing, and controls hardware included in the computer. The processor 11 is, for example, a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or a GPU (Graphics Processing Unit).
The inference device 100 may include a plurality of processors that replace the processor 11. A plurality of processors share the role of the processor 11.

Memory 12 is typically a volatile storage device. Memory 12 is also called main storage or main memory. The memory 12 is, for example, a RAM (Random Access Memory). The data stored in the memory 12 is stored in the auxiliary storage device 13 as necessary.

The auxiliary storage device 13 is typically a nonvolatile storage device. The auxiliary storage device 13 is, for example, a ROM (Read Only Memory), an HDD (Hard Disk Drive), or a flash memory. Data stored in the auxiliary storage device 13 is loaded into the memory 12 as needed.
The memory 12 and the auxiliary storage device 13 may be configured integrally.

The input/output IF 14 is a port to which an input device and an output device are connected. The input/output IF 14 is, for example, a USB (Universal Serial Bus) terminal. Specific examples of input devices include a camera, a keyboard, and a mouse. A specific example of the output device is a display.

Communication device 15 is a receiver and a transmitter. The communication device 15 is, for example, a communication chip or a NIC (Network Interface Card).

Each part of the inference device 100 may use the communication device 15 as appropriate when communicating with other devices. Each part of the inference device 100 may receive data via the input/output IF 14 or may receive data via the communication device 15.

The auxiliary storage device 13 stores an inference program. The inference program is a program that causes a computer to realize the functions of each part of the inference device 100. The inference program is loaded into memory 12 and executed by processor 11. The functions of each part included in the inference device 100 are realized by software.
Further, the auxiliary storage device 13 stores an OS (Operating System). At least a portion of the OS is loaded into memory 12 and executed by processor 11. That is, the processor 11 executes the inference program while executing the OS.

Data used when executing the inference program, data obtained by executing the inference program, etc. are appropriately stored in the storage device. Each part of the inference device 100 uses a storage device as appropriate. The storage device includes, as a specific example, at least one of the memory 12, the auxiliary storage device 13, a register within the processor 11, and a cache memory within the processor 11. Note that data and information may have the same meaning. The storage device may be independent of the computer.
The functions of the memory 12 and the auxiliary storage device 13 may be realized by other storage devices.

The inference program may be recorded on a computer-readable non-volatile recording medium. Specific examples of the nonvolatile recording medium include an optical disk or a flash memory. The reasoning program may be provided as a program product.

***Operation explanation***
The operation procedure of the inference device 100 corresponds to an inference method. Further, a program that realizes the operation of the inference device 100 corresponds to an inference program.

FIG. 4 is a flowchart illustrating an example of the operation of the inference device 100. Further, FIG. 5 is a diagram schematically explaining an example of the operation of the inference device 100. In FIG. 5, parallelograms represent data used in calculations, excluding output data. The operation of the inference device 100 will be explained with reference to FIGS. 4 and 5. Note that this embodiment can be applied to a machine learning method that can also use quantized data; however, for convenience of explanation, the machine learning method in the explanation of the operation of the inference device 100 is based on deep learning. shall be.

(Step S101)
The quantization inference process of the quantization inference unit 110 is activated.

(Step S102)
If there is an instruction to start inference, the inference device 100 proceeds to step S103. In other cases, the inference device 100 executes the process of this step again.

(Step S103)
The quantization inference unit 110 starts inference processing using the inference data. The inference data may be data received by the inference device 100 together with an instruction to start inference.

(Step S104)
The quantization inference unit 110 performs layer calculations on the target layer. Here, the layer is a layer in deep learning, and in the loop processing consisting of steps S104 to S106, when layer calculation is executed for the first time, the first layer is the target layer, and in other cases, the layer that was executed one time ago is the target layer. The layer next to the target layer in the calculation is the target layer.
The calculations in FIG. 5 correspond to layer calculations. Further, the quantization inference unit 110 executes a layer operation using the data shown on the left side of the thick arrow shown directly below the operation as input data, and outputs the data shown on the right side of the arrow. .

(Step S105)
The quantization inference unit 110 checks whether an overflow has occurred in the process of step S104. Examples of ways to check whether an overflow has occurred are to refer to CPU flags, to check variables related to calculations before and after calculations, or to check using a unique circuit such as an FPGA. It's a method.
If an overflow occurs, the inference device 100 proceeds to step S109. Otherwise, the inference device 100 proceeds to step S106.

(Step S106)
When all layer calculations are completed, the inference device 100 proceeds to step S107. Otherwise, the inference device 100 proceeds to step S104.

(Step S107)
The quantization inference unit 110 appropriately notifies the person who instructed the inference of the result of the inference.

(Step S108)
If an overflow occurs in at least one of the layer calculations, the inference device 100 proceeds to step S111. Otherwise, the inference device 100 returns to step S102.

(Step S109)
The data extraction unit 130 saves the save data corresponding to the target layer. As a specific example, the data extraction unit 130 extracts data regarding the target layer and saves the extracted data as save data. Specific examples of the data regarding the target layer include input data for the target layer, variable data in the target layer, and feature data in the target layer. However, if the target layer is the first layer, the data extraction unit 130 does not extract the input data as saved data because the input data is inference data. Note that the data extraction unit 130 may save the inference data. Here, the inference data is data input to the first layer.
FIG. 5 shows how the data extraction unit 130 saves data about the target layer where an overflow has occurred and data about the layer immediately before the target layer, as saved data corresponding to the target layer. Note that the data saved by the data extraction unit 130 may be only data about the target layer, or may be data about each layer from the target layer to the layer n (n is a natural number) before the target layer. It's okay.

(Step S110)
The quantization inference unit 110 performs a saturation operation, and then proceeds to step S106.

(Step S111)
The non-quantized inference process of the non-quantized inference unit 120 is activated.
Hereinafter, the inference device 100 executes the processing after step S102 and the processing after step S121 in parallel.

(Step S121)
The non-quantized inference unit 120 executes a non-quantized inference process using the inference data.
Note that in FIG. 5, the non-quantized inference unit 120 outputs output data by executing the non-quantized inference process to the end, but the non-quantized inference unit 120 outputs output data when an overflow occurs in the quantized inference process. It is sufficient to perform layer calculations up to the layer corresponding to the layer.

(Step S122)
The data extraction unit 130 extracts, as saved data, data regarding a layer corresponding to the layer in which the overflow has occurred. Here, the data extraction unit 130 saves data regarding the same layer as the layer corresponding to the data saved in the quantization inference process. As a specific example, as shown in FIG. 5, when the data extraction unit 130 saves data for each of the layer in which an overflow occurred in the quantization inference process and the layer immediately before the layer, the data extraction unit 130 saves, as save data, data regarding the layer of the non-quantized inference process corresponding to each of the layer corresponding to the layer in which the overflow has occurred and the layer immediately before the layer in which the overflow has occurred.
The data in the quantized inference process and the data in the non-quantized inference process saved by the data extraction unit 130 may be output to the outside of the inference device 100. Engineers, etc. may obtain the output data and reset the parameters of the quantization inference process based on the obtained data, and may change the parameters related to the quantization inference process in the source code based on the obtained data. You may.

(Step S123)
The non-quantization inference process of the non-quantization inference unit 120 is terminated.

***Explanation of effects of Embodiment 1***
As described above, according to the present embodiment, even if the inference data is confidential information, the inference data is not extracted, but only the data related to the operation in which the overflow occurred. Here, the data related to the operation in which the overflow has occurred is data from which the confidentiality that exists in the inference data has been eliminated. Therefore, according to the present embodiment, it is possible to analyze an overflow that occurs in a quantization operation while maintaining accuracy in the inference environment and maintaining confidentiality. Therefore, according to the present embodiment, it is possible to obtain data necessary for improving the trained model to deal with the overflow that has occurred.
Furthermore, according to the present embodiment, parameters and the like in machine learning involving quantization can be adjusted based not only on learning data and the like but also on actual inference data.

***Other configurations***
<Modification 1>
The quantization inference unit 110 may terminate inference midway if overflow occurs. As a specific example, if an overflow occurs in at least one quantization operation, the quantization inference unit 110 does not execute the operation subsequent to the operation in which the overflow occurred.

<Modification 2>
FIG. 6 shows an example of the hardware configuration of the inference device 100 according to this modification. The inference device 100 according to this modification includes an offload device 16, as shown in the figure.
The quantization inference process may be performed by offload device 16. The offload device 16 is, for example, a GPU or an FPGA. In this modification, the non-quantized inference process may be executed by the processor 11, and at this time, communication is executed between the processor 11 and the offload device 16 as appropriate.

<Modification 3>
FIG. 7 shows an example of the hardware configuration of the inference device 100 according to this modification.
The inference device 100 includes a processing circuit 18 in place of the processor 11, the processor 11 and memory 12, the processor 11 and auxiliary storage 13, or the processor 11, memory 12, and auxiliary storage 13.
The processing circuit 18 is hardware that implements at least a portion of each unit included in the inference device 100.
Processing circuit 18 may be dedicated hardware or may be a processor that executes a program stored in memory 12.

When the processing circuit 18 is dedicated hardware, the processing circuit 18 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (ASIC stands for Application Specific Integrated Circuit), or an FPGA. (Field Programmable Gate Array) or a combination thereof.
The inference device 100 may include a plurality of processing circuits that replace the processing circuit 18. The plurality of processing circuits share the role of the processing circuit 18.

In the inference device 100, some functions may be realized by dedicated hardware, and the remaining functions may be realized by software or firmware.

The processing circuit 18 is implemented, for example, by hardware, software, firmware, or a combination thereof.
The processor 11, memory 12, auxiliary storage device 13, and processing circuit 18 are collectively referred to as a "processing circuitry." That is, the functions of each functional component of the inference device 100 are realized by processing circuitry.
The inference device 100 according to other embodiments may also have the same configuration as this modification.

Embodiment 2.
Hereinafter, differences from the embodiments described above will be mainly described with reference to the drawings.

***Explanation of configuration***
FIG. 8 shows a configuration example of the inference device 100 according to this embodiment. The inference device 100 includes a requantization unit 140 as shown in the figure.
The requantization unit 140 generates a requantization algorithm by changing the quantization algorithm based on the data extracted by the data extraction unit 130. Here, the requantization algorithm is a quantization algorithm modified by the requantization unit 140. Further, the requantization unit 140 manages the timing to perform requantization. Requantization is changing the requantization algorithm, and may include replacing the quantization algorithm with a requantization algorithm. Based on the quantized feature data and non-quantized feature data extracted by the data extraction unit 130, the requantization unit 140 determines whether overflow occurs in the target operation, which is a quantization operation corresponding to the extracted quantized feature data. Change the quantization feature data corresponding to the target operation so that this does not occur.
The quantization inference unit 110 according to the present embodiment appropriately utilizes the requantization algorithm generated by the requantization unit 140 and executes a quantization operation according to the changed quantization feature data. Further, the quantization inference unit 110 manages the timing of replacing the re-quantization algorithm with the currently operating quantization algorithm.

***Operation explanation***
FIG. 9 is a flowchart showing an example of the operation of the inference device 100. With reference to this figure, differences between the operation of inference device 100 according to the first embodiment and the operation of inference device 100 according to the present embodiment will be mainly explained.

(Step S102)
The inference device 100 proceeds to step S201 instead of executing the process of this step again.

(Step S108)
The inference device 100 proceeds to step S201 instead of returning to step S102.

(Step S201)
If the requantization algorithm is not prepared by the requantization unit 140, the quantization inference unit 110 proceeds to step S102. In other cases, the quantization inference unit 110 proceeds to step S202.

(Step S202)
The quantization inference unit 110 replaces the quantization algorithm in the quantization inference process with the requantization algorithm prepared by the requantization unit 140.

(Step S221)
The requantization unit 140 generates a requantization algorithm by performing requantization using the saved data. Requantization is adjusting the quantization algorithm so that overflow does not occur in the layer where overflow has occurred. As a specific example, the requantization unit 140 changes the feature data in the layer where the overflow has occurred.
Note that when overflow occurs in multiple layers, the requantization unit 140 may adjust only the feature data corresponding to the layer that is processed earliest, and may adjust the feature data corresponding to each of all the layers in which overflow has occurred. You may adjust all the feature data at once.

(Step S222)
The requantization unit 140 stores the requantization algorithm generated in step S221.

***Explanation of effects of Embodiment 2***
As described above, according to the present embodiment, it is possible to automatically execute the process of changing the quantization algorithm so that overflow does not occur.

***Other configurations***
<Modification 4>
The timing at which the quantization inference unit 110 replaces the quantization algorithm in the quantization inference process with the quantization algorithm prepared by the requantization unit 140 may be when the embedded system including the inference device 100 is restarted. In addition, the quantization inference unit 110 replaces the quantization algorithm by a method that takes into account the timing of the next inference and makes an operation reservation to replace the quantization algorithm when it is determined that the next inference will not be affected. It's okay.

***Other embodiments***
It is possible to freely combine the embodiments described above, to modify any component of each embodiment, or to omit any component in each embodiment.
Further, the embodiments are not limited to those shown in Embodiments 1 and 2, and various changes can be made as necessary. The procedures described using flowcharts and the like may be modified as appropriate.

11 processor, 12 memory, 13 auxiliary storage device, 14 input/output IF, 15 communication device, 16 offload device, 18 processing circuit, 19 signal line, 100 inference device, 110 quantization inference section, 120 non-quantization inference section, 130 data extraction section, 140 requantization section.

Claims (6)

a quantization inference unit that performs at least one quantization operation based on a machine learning method using the inference data;
An inference device comprising: a non-quantization inference unit that executes at least one of at least one non-quantization operation corresponding to each of the at least one quantization operation using the inference data,
Each of the at least one quantization operation is an operation corresponding to at least one quantization feature data indicating a characteristic of each of the at least one quantization operation,
Each of the at least one non-quantized operation is an operation according to each of at least one non-quantized feature data indicating the characteristics of each of the at least one non-quantized operation,
The inference device further includes:
When an overflow occurs in at least one of the at least one quantization operation, quantization feature data corresponding to each quantization operation in which an overflow occurred, and quantization feature data corresponding to each quantization operation in which an overflow occurred. An inference device comprising a data extraction unit that extracts non-quantized feature data corresponding to a non-quantized operation. Each of the at least one quantization feature data includes a parameter corresponding to each of the at least one quantization operation,
The inference device according to claim 1, wherein each of the at least one non-quantized feature data includes a parameter corresponding to each of the at least one non-quantized operation. The machine learning method is deep learning,
The data extraction unit is
Extracting data indicating parameters for a layer corresponding to a quantization operation in which an overflow has occurred as the quantization feature data;
The inference device according to claim 1 or 2, wherein data indicating a parameter for a layer corresponding to a non-quantized operation corresponding to a quantized operation in which an overflow has occurred is extracted as the non-quantized feature data. The inference device further includes:
Based on the extracted quantized feature data and non-quantized feature data, quantization corresponding to the target operation is performed to prevent overflow from occurring in the target operation, which is a quantization operation corresponding to the extracted quantized feature data. Equipped with a requantization unit that changes feature data,
The inference device according to any one of claims 1 to 3, wherein the quantization inference unit executes a quantization operation according to changed quantized feature data. the computer executes at least one quantization operation based on a machine learning method using the inference data;
An inference method , wherein the computer uses the inference data to execute at least one of at least one non-quantization operation corresponding to each of the at least one quantization operation,
Each of the at least one quantization operation is an operation corresponding to at least one quantization feature data indicating a characteristic of each of the at least one quantization operation,
Each of the at least one non-quantized operation is an operation according to each of at least one non-quantized feature data indicating the characteristics of each of the at least one non-quantized operation,
When an overflow occurs in at least one of the at least one quantization operation, the computer stores quantization feature data corresponding to each of the quantization operations in which the overflow has occurred, and quantization feature data for each of the quantization operations in which the overflow has occurred. An inference method for extracting non-quantized feature data corresponding to each corresponding non-quantized operation. quantization inference processing that performs at least one quantization operation based on a machine learning method using inference data;
An inference that causes an inference device, which is a computer, to perform a non-quantized inference process of executing at least one of at least one non-quantized operation corresponding to each of the at least one quantized operation using the inference data. A program,
Each of the at least one quantization operation is an operation corresponding to at least one quantization feature data indicating a characteristic of each of the at least one quantization operation,
Each of the at least one non-quantized operation is an operation according to each of at least one non-quantized feature data indicating the characteristics of each of the at least one non-quantized operation,
The inference program further includes:
When an overflow occurs in at least one of the at least one quantization operation, quantization feature data corresponding to each quantization operation in which an overflow occurred, and quantization feature data corresponding to each quantization operation in which an overflow occurred. An inference program that causes the inference device to execute a data extraction process of extracting non-quantized feature data corresponding to a non-quantized operation.

Images