# US 2021/0081802 A1 · Information processing device and method, and recording medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2021/0081802 A1 - Document ID: US20210081802A1 - Status: Published application - Kind: A1 - Application number: 16/997,989 - Filed: 2020-08-20 - Published: 2021-03-18 - Earliest claimed: 2019-09-13 - Assignees: Fujitsu Limited - Applicants: Fujitsu Limited - Inventors: Yasufumi Sakai - Web page: https://patentgrep.com/patent/US20210081802A1 ## CPC - G06N 3/084: Backpropagation, e.g. using gradient descent - G06F 7/49942: {Significance control} - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/0495: Quantised networks; Sparse networks; Compressed networks - G06N 3/063: using electronic means - G06N 3/09: Supervised learning - G06N 5/046: Forward inferencing; Production systems - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract An information processing device, includes a memory; and a processor coupled to the memory and configured to: calculate a quantization error when a variable to be used in a neural network is quantized, generate a threshold value based on reference information related to a first recognition rate obtained by past learning of the neural network and a second recognition rate that is obtained by calculation of the neural network, determine a variable of data type to be quantized among variables to be used for calculation of the neural network based on the calculated quantization error and the generated threshold value, and execute the calculation of the neural network by using the variable of data type. ## Claims ### Claim 1 An information processing device, comprising: a memory; and a processor coupled to the memory and configured to: calculate a quantization error when a variable to be used in a neural network is quantized, generate a threshold value based on reference information related to a first recognition rate obtained by past learning of the neural network and a second recognition rate that is obtained by calculation of the neural network, determine a variable of data type to be quantized among variables to be used for calculation of the neural network based on the calculated quantization error and the generated threshold value, and execute the calculation of the neural network by using the variable of data type. ### Claim 2 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to execute calculation of the neural network by dividing the calculation into a plurality of calculation cycles, and in each of the plurality of calculation cycles, a process for executing the calculation of the neural network includes: In a type determination period in which a variable to be quantized is determined, executing the calculation of the neural network by using a variable of a data type determined in previous one of the calculation cycles, and in a calculation execution period after the type determination period, executing the calculation of the neural network by using the variable of the determined data type. ### Claim 3 (depends on 2) The information processing device according to claim 2, wherein the process for calculating a quantization error, the process for generating a threshold value, and the process for determining a quantization variable are performed in the type determination period, and are not executed in the calculation execution period. ### Claim 4 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to: calculate a first difference between the first recognition rate indicated by the reference information and the second recognition rate, generate an update amount of the threshold value based on the calculated first difference, and calculate a threshold value after updating based on the generated update amount and the current threshold value. ### Claim 5 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to: generate a threshold value includes generating the single threshold value, and determine a variable to be quantized includes determining the variable to be quantized among all the variables based on the generated common threshold value. ### Claim 6 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to: generate a threshold value includes generating the threshold value for each type of the variables, and determine a variable to be quantized includes determining the variable to be quantized for each type of the variables based on the generated threshold value for each type. ### Claim 7 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to: determine a data type of a first variable among a plurality of variables to be used in the neural network based on the quantization error and the threshold value, and fix a data type of a second variable other than the first variable to a fixed-point type. ### Claim 8 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to calculate the quantization error of a variable to be used in each layer for each of a plurality of layers included in the neural network. ### Claim 9 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to determine the variable to be quantized by a unit of the variables to be used in each layer based on the calculated quantization error and the generated threshold value. ### Claim 10 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to execute learning of the neural network by using the variable of the determined data type. ### Claim 11 (depends on 1) The information processing device according to claim 1, wherein the processor is configured to execute inference of the neural network by using the variable of the determined data type. ### Claim 12 An information processing method executed by a computer, the information processing method comprising: calculating a quantization error when a variable to be used in a neural network is quantized, generating a threshold value based on reference information related to a first recognition rate obtained by past learning of the neural network and a second recognition rate that is obtained by calculation of the neural network, determining a variable of data type to be quantized among variables to be used for calculation of the neural network based on the calculated quantization error and the generated threshold value, and executing the calculation of the neural network by using the variable of data type. ### Claim 13 A non-transitory computer-readable storage medium storing a program that causes a computer to execute a process, the process comprising: calculating a quantization error when a variable to be used in a neural network is quantized, generating a threshold value based on reference information related to a first recognition rate obtained by past learning of the neural network and a second recognition rate that is obtained by calculation of the neural network, determining a variable of data type to be quantized among variables to be used for calculation of the neural network based on the calculated quantization error and the generated threshold value, and executing the calculation of the neural network by using the variable of data type.