Information processing apparatus and information processing method
AssigneeRakuten Group, Inc.
InventorsAlois De La Comble, Ken Prepin
Life of this patent
Estimating the term…
Abstract
Disclosed herein is an information processing apparatus including: a training data acquisition unit configured to acquire training data; an input channel setting unit configured to set a plurality of input channels for inputting different features, respectively, from the training data acquired by the training data acquisition unit to a learning model to be trained with the training data; and a training unit configured to, in each training, randomly select, among the plurality of input channels set by the input channel setting unit, at least a part of the input channels, and deactivate an input from the selected input channel to the learning model to repeat training of the learning model.
Claims14
An information processing apparatus, comprising:
- at least one memory configured to store program code;
- at least one processor configured to operate as instructed by the program code, the program code including:
- acquisition code configured to cause at least one of the at least one processor to acquire training data;
- setting code configured to cause at least one of the at least one processor to set a plurality of input channels for inputting different features from the training data acquired—to a learning model to be trained with the training data and organize the plurality of input channels into a first input channel group for inputting features associated with color from the training data and a second input channel group for inputting features associated with text from the training data;
- training code configured to cause at least one of the at least one processor to, in each training, randomly select, among the plurality of input channels set, at least a part of the input channels, deactivate an input channel from the selected input channel to the learning model to repeat training of the learning model, and rescale remaining input channels by a factor equal to a total number of the plurality of input channels divided by a number of remaining active input channels to maintain constant input magnitude to the learning model;
- and
- inferencing code configured to cause at least one of the at least one processor to activate all of the plurality of input channels set with respect to the learning model trained, and input input data to the learning model to perform inference.
An information processing method, performed by at least one processor and comprising:
- acquiring training data;
- setting a plurality of input channels for inputting different features, respectively, from the training data to a learning model to be trained with the training data;
- organizing the plurality of input channels into a first input channel group for inputting features associated with color from the training data and a second input channel group for inputting features associated with text from the training data;
- in each training, randomly selecting, among the plurality of input channels, at least a part of the input channels, and deactivating an input from the selected input channel to the learning model to repeat training of the learning model;
- rescaling remaining input channels by a factor equal to a total number of the plurality of input channels divided by a number of remaining active input channels to maintain constant input magnitude to the learning model;
- and
- activate all of the plurality of input channels with respect to the learning model, and input input data to the learning model to perform inference.
A non-transitory computer readable medium storing a computer program that when executed by at least one processor, causes the at least one processor to:
- acquire training data;
- set a plurality of input channels for inputting different features, respectively, from the training data to a learning model to be trained with the training data;
- organize the plurality of input channels into a first input channel group for inputting features associated with color from the training data and a second input channel group for inputting features associated with text from the training data;
- in each training, randomly select, among the plurality of input channels, at least a part of the input channels, and deactivate an input from the selected input channel to the learning model to repeat training of the learning model;
- rescale remaining input channels by a factor equal to a total number of the plurality of input channels divided by a number of remaining active input channels to maintain constant input magnitude to the learning model;
- and
- activate all of the plurality of input channels with respect to the learning model, and input input data to the learning model to perform inference.
Dependencies are inferred from the claim text.