Model inference method based on incremental learning and electronic device

Abstract

The application relates to a model inference method based on incremental learning and an electronic device. The method comprises the following steps: performing a clustering operation on each batch of training sample sets respectively, constructing a template sample library corresponding to each batch of training sample sets according to the clustering results, training an initial model by using each batch of training sample sets to obtain a target model corresponding to each batch of training sample sets, obtaining a test sample set, calculating a similarity value between each test sample in the test sample set and each template sample in each template sample library, determining a target template sample library corresponding to a template sample with a maximum value in the similarity value, and performing inference on the test sample set by using a target model corresponding to the target template sample library to obtain an inference result of each test sample. The application can prevent the problem of catastrophic forgetting and improve the accuracy of inference on the test sample set.

Images