| CPC G06N 5/04 (2013.01) [G06F 18/217 (2023.01); G06F 18/2431 (2023.01)] | 15 Claims |
|
1. A learning device comprising:
a memory storing software instructions; and
at least one processor configured to execute the software instructions to perform:
storing, in a predictor storage, one existing predictors that have already been learned for seen tasks or seen classes via one-shot learning, each existing predictor having a corresponding seen task or a corresponding seen class;
for each existing predictor, storing an attribute vector of the corresponding seen task or the corresponding seen class, in an attribute storage;
for each existing predictor, calculating an prediction score for test input data, wherein the prediction score indicates a likelihood the test input data is the corresponding seen task of the existing predictor or belongs to the corresponding seen class of the existing predictor;
for each existing predictor, inferring correspondence between an attribute vector of an unseen task or an unseen class and the corresponding attribute vector or the corresponding seen class of the existing predictor, wherein the unseen task or the unseen class is a task or class for which an existing predictor has not already been learned;
calculating a prediction score for the test input data for the unseen task or the unseen class indicating a likelihood that the test input data is the unseen task or belongs to the unseen class, based on the calculated prediction score and the inferred correspondence for each existing predictor; and
outputting the calculated prediction score indicating the likelihood that the test input data is the unseen task or belongs to the unseen class.
|