Performing distributed processing using layers of a neural network divided between a first device and a second device
An information processing method according to the present disclosure includes the steps of: evaluating, by a computer, a neural network having a structure held in a divided manner by a first device and a second device based on information on transfer of information between the first device and the second device in the neural network; and determining, by the computer, the structure of the neural network based on the evaluation of the neural network.
1 . An information processing method executed by processing circuitry, the method comprising:
evaluating transfer of information between layers of a neural network performing distributed processing, wherein the layers of the neural network are divided between a first device and a second device, to determine an evaluation value;
wherein:
a third layer among the layers of the neural network that lies in a portion deeper than a second layer from which second information having a maximum size is output, the third layer outputs third information having a size smaller than that of first information output from an input layer of the neural network, the third layer is determined to be a transfer point at which the third information is to be transmitted from the first device to the second device, and
the neural network is evaluated based on the third information on the determined transfer point;
determining if the evaluation value meets a predetermined condition;
determining a structure of the neural network that meets the predetermined condition; and
executing a distributed process using the determined structure in which the layers of the neural network are divided between the first device and the second device.
2 . The information processing method according to claim 1 , wherein the neural network is evaluated based on a number of layers that lie in portions shallower than the transfer point and a total number of the layers constituting the neural network.
3 . The information processing method according to claim 1 , wherein the neural network is evaluated based on an indicator value that represents a recognition performance of the neural network.
4 . The information processing method according to claim 1 , wherein the neural network is evaluated based on a calculation amount in the neural network.
5 . The information processing method according to claim 1 , wherein the neural network is evaluated based on information on a performance of arithmetic processing of the first device.
6 . The information processing method according to claim 5 , wherein the neural network is evaluated based on a number of times of floating-point operations and a number of times of operations other than the floating-point operations in each layer of the neural network held in the first device.
7 . The information processing method according to claim 5 , wherein the neural network is evaluated based on a relation between a number of times of multiplication and a number of times of operations other than the multiplication performed in each layer of the neural network held in the first device.
8 . The information processing method according to claim 1 , wherein the neural network is evaluated based on values obtained by multiplying the information on transfer, an indicator value that represents a recognition performance of the neural network, a calculation amount in the neural network, and information on a performance of arithmetic processing of the first device by respective predetermined weight values.
9 . The information processing method according to claim 8 , wherein the weight values are determined based on configurations of the first device and the second device, a communication standard between the first device and the second device, and information on an environment in which the neural network is provided.
10 . An information processing device comprising:
processing circuitry configured to:
evaluate transfer of information between layers of a neural network performing distributed processing, wherein the layers of the neural network are divided between a first device and a second device, to determine an evaluation value;
wherein a third layer among the layers of the neural network that lies in a portion deeper than a second layer from which second information having a maximum size is output, the third layer outputs third information having a size smaller than that of first information output from an input layer of the neural network, the third layer is determined to be a transfer point at which the third information is to be transmitted from the first device to the second device, and
the neural network is evaluated based on the third information on the determined transfer point;
determine if the evaluation value meets a predetermined condition;
determine a structure of the neural network that meets the predetermined condition; and
execute a distributed process using the determined structure in which the layers of the neural network are divided between the first device and the second device.
11 . A non-transitory computer readable medium storing instructions that, when executed by processing circuitry, perform an information processing method comprising:
evaluating transfer of information between layers of a neural network performing distributed processing, wherein the layers of the neural network are divided between a first device and a second device, to determine an evaluation value;
wherein:
a third layer among the layers of the neural network that lies in a portion deeper than a second layer from which second information having a maximum size is output, the third layer outputs third information having a size smaller than that of first information output from an input layer of the neural network, the third layer is determined to be a transfer point at which the third information is to be transmitted from the first device to the second device, and
the neural network is evaluated based on the third information on the determined transfer point;
determining if the evaluation value meets a predetermined condition;
determining a structure of the neural network that meets the predetermined condition; and
executing a distributed process using the determined structure in which the layers of the neural network are divided between the first device and the second device.