US 11,537,930 B2Granted

Information processing device, information processing method, and program

AssigneeNEC Corporation

InventorAtsushi Sato

Life of this patent

Estimating the term…

Loading patent…

Abstract

An information processing device which performs semi-supervised learning is provided with: a dictionary input circuit for acquiring a dictionary, a parameter group used by an identification device; a boundary determination circuit which obtains an identification boundary for the dictionary on the basis of the dictionary, supervised data, and labelled unsupervised data; a labelling circuit which labels the unsupervised data in accordance with the identification boundary; a loss calculation circuit which calculates the sum total of supervised-data loss calculated in accordance with the labels assigned in advance and the labels based on the identification boundary, and unsupervised-data loss calculated such that further from the identification boundary the smaller the loss; a dictionary update circuit which updates the dictionary such that the sum-total loss is reduced; and a dictionary output circuit which outputs the updated dictionary.

Claims17

Dependent claims

An information processing device which uses supervised data and unsupervised data to perform semi-supervised learning, the information processing device comprising:

at least one memory configured to store instructions, and;
at least one processor configured to execute instructions to:
obtain a dictionary that includes a parameter group used in an identification device;
search for an identification boundary of the dictionary based on the dictionary and the supervised data;
set a first loss for evaluating a loss of the supervised data and the unsupervised data;
assign labels to samples of the unsupervised data in accordance with the identification boundary;
calculate a loss of the supervised data as a supervised loss by applying a function to the supervised data when a preassigned label which is preassigned to the supervised data is different from a determined label determined based on the identification boundary;
calculate a loss of the unsupervised data as an unsupervised loss by giving a predetermined value of loss only to the samples that are within a certain range from the identification boundary among the samples labeled in the unsupervised data;
calculate a sum of the supervised loss and the unsupervised loss, as a second loss;
when the second loss is less than the first loss,
update the dictionary to shift the identification boundary;
replace the first loss with the calculated second loss;
search for a new identification boundary of the dictionary based on the dictionary, the supervised data, and the unsupervised data;
repeat processing from assignment of the labels to the samples of the unsupervised data in accordance with the identification boundary;
and
when the second loss is not less than the first loss, output the updated dictionary.

13 more elements

An information processing method executed by an information processing device which uses supervised data and unsupervised data to perform semi-supervised learning, the information processing method comprising:

obtaining a dictionary which includes a parameter group used in an identification device;
setting a first loss for evaluating a loss of the supervised data and the unsupervised data;
searching an identification boundary of the dictionary based on the dictionary and the supervised data;
assigning labels to samples of the unsupervised data in accordance with the identification boundary;
calculating a loss of the supervised data as a supervised loss by applying a function to the supervised data when a preassigned label which is preassigned to the supervised data is different from a determined label determined based on the identification boundary;
calculating a loss of the unsupervised data as an unsupervised loss by giving predetermined value of loss only to the samples that are within a certain range from the identification boundary among the samples labeled in the unsupervised data, as the unsupervised loss;
calculating a sum of the supervised loss and the unsupervised loss, as a second loss;
when the second loss is less than the first loss,
updating the dictionary to shift the identification boundary;
replacing the first loss with the calculated second loss;
searching for a new identification boundary of the dictionary based on the dictionary, the supervised data, and the unsupervised data;
repeating from assigning the labels to the samples of the unsupervised data in accordance with the identification boundary;
and
when the second loss is not less than the first loss, outputting the updated dictionary.

11 more elements

A non-transitory computer readable medium storing a program executed by an information processing device which uses supervised data and unsupervised data to perform semi-supervised learning, the program causing the information processing device to execute:

obtaining a dictionary which includes a parameter group used in an identification device;
setting a first loss for evaluating a loss of the supervised data and the unsupervised data;
searching an identification boundary of the dictionary based on the dictionary and the supervised data;
assigning labels to samples of the unsupervised data in accordance with the identification boundary;
calculating a loss of the supervised data as a supervised loss by applying a function to the supervised data when a preassigned label which is preassigned to the supervised data is different from a determined label determined based on the identification boundary;
calculating a loss of the unsupervised data as an unsupervised loss by giving predetermined value of loss only to the samples that are within a certain range from the identification boundary among the samples labeled in the unsupervised data, as the unsupervised loss;
calculating a sum of the supervised loss and the unsupervised loss, as a second loss;
when the second loss is less than the first loss,
updating the dictionary to shift the identification boundary;
replacing the first loss with the calculated second loss;
searching for a new identification boundary of the dictionary based on the dictionary, the supervised data, and the unsupervised data;
repeating from assigning the labels to the samples of the unsupervised data in accordance with the identification boundary;
and
when the second loss is not less than the first loss, outputting the updated dictionary.

11 more elements

Dependencies are inferred from the claim text.