A METHOD AND SYSTEM OF SWITCHING SPECIALIZED LIGHTWEIGHT NEURAL NETWORKS FOR CONCEPT-DRIFT ADAPTATION IN RESOURCE-CONSTRAINED MOBILE DEVICES

摘要

A specialized lightweight neural network replacement method and apparatus for adapting to dynamic data transformation in resource-constrained mobile devices are disclosed. A neural network replacement method according to an embodiment includes: constructing a model database by creating a specialized neural network for each task classified based on semantic similarity; and performing inference by replacing it with a specialized neural network corresponding to a task related to a user's inference request based on the built model database.