A Method And A System For Adapting Neural Network To Data Drifts

Abstract: “A METHOD AND A SYSTEM FOR ADAPTING NEURAL NETWORK TO DATA DRIFTS” ABSTRACT The present disclosure relates to adapting neural network (102) to data drifts. The method comprises receiving training input data and plurality of classes, associated with at least one task of pre-trained neural network (102); identifying mapping between a plurality of weights and plurality of classes, based on training input data; identifying expected types of data drifts associated with training input data, based on plurality of pre-defined types of data drifts; generating sets of updated weights corresponding to each of plurality of classes; and providing sets of updated weights corresponding to each of plurality of classes to neural network (102), to perform at least one task on occurrence of expected types of data drifts in input data. Therefore, the neural network (102) can adapt to data drifts in input data, thereby increasing the accuracy of the neural network. FIGURE 1

Patent Information

Applicants

Inventors

Specification

Documents

Application Documents