Information processing method and information processing system
View Patent ↗An information processing method including the following executed using a computer: obtaining a neural network model that solves a regression problem; obtaining input data and label data corresponding to the input data; compressing a network of the neural network model to obtain a compressed model; transforming the regression problem to be solved by the neural network model into a classification problem, based on information indicating performance of the compressed model, the information being calculated using the label data and output data which is obtained by inputting the input data to the compressed model; and changing a network configuration of the neural network model and transforming the label data, in accordance with the transformation from the regression problem to the classification problem.
1. An information processing method, comprising the following executed using a processor and a memory:
storing, in the memory, a neural network model that solves a regression problem, input data, and label data corresponding to the input data,
obtaining, by the processor, the neural network model, the input data, and the label data from the memory;
compressing, by the processor, a network of the neural network model to obtain a compressed model;
transforming, by the processor, the regression problem to be solved by the neural network model into a classification problem when accuracy of output data which is obtained by inputting the input data to the compressed model is lower than or equal to a threshold, the accuracy being calculated using the label data and the output data;
changing, by the processor, a network configuration of the neural network model in accordance with the transformation from the regression problem to the classification problem; and
transforming, by the processor, label data corresponding to the regression problem to label data corresponding to the classification problem.
2. The information processing method according to claim 1 , wherein
the changing of the network configuration includes changing the number of nodes in the neural network model.
3. The information processing method according to claim 2 , wherein
the changing of the number of nodes includes increasing the number of nodes in an output layer of the neural network model.
4. The information processing method according to claim 3 , wherein
the increasing of the number of nodes includes changing a distribution of the nodes.
5. The information processing method according to claim 1 , wherein
the changing of the network configuration includes changing the number of nodes in the neural network model within a range in which a condition for a computational resource is satisfied, the computational resource being usable by the compressed model.
6. The information processing method according to claim 1 , wherein the information indicating the accuracy is information based on a difference between the output data and the label data.
7. The information processing method according to claim 1 , wherein
the compressing of the network includes changing a weight for the neural network model in accordance with a change in computation precision from floating-point to fixed-point representation.
8. The information processing method according to claim 1 , wherein
the changing of the network configuration includes setting output-weights each being set for a different one of outputs of nodes in the neural network model, and
an output of the neural network model is determined using the outputs and the output-weights of the nodes.
9. An information processing system, comprising:
a memory configured to store a program; and
a processor configured to execute the program and control the information processing system to:
store, in the memory, a neural network model that solves a regression problem, input data, and label data corresponding to the input data,
obtain the neural network model, the input data, and the label data from the memory;
compress a network of the neural network model to obtain a compressed model;
transform the regression problem to be solved by the neural network model into a classification problem when accuracy of output data which is obtained by inputting the input data to the compressed model is lower than or equal to a threshold, the accuracy being calculated using the label data and the output data;
change a network configuration of the neural network model in accordance with the transformation from the regression problem to the classification problem; and
transform label data corresponding to the regression problem to label data corresponding to the classification problem.
10. An information processing method, comprising the following executed using a processor and a memory:
storing, in the memory, a neural network model for object detection which solves a regression problem, input image data and label data corresponding to the input image data;
obtaining, by the processor, the neural network model, the input image data, and the label data from the memory;
compressing, by the processor, a network of the neural network model to obtain a compressed model;
transforming, by the processor, the regression problem to be solved by the neural network model into a classification problem when accuracy of output data which is obtained by inputting the input image data to the compressed model is lower than or equal to a threshold, the accuracy being calculated using the label data and the output data
changing, by the processor, a network configuration of the neural network model in accordance with the transformation from the regression problem to the classification problem; and
transforming, by the processor, label data corresponding to the regression problem to label data corresponding to the classification problem.