(A) pragmatic approach to on-device incremental learning system with selective weight updates

Abstract

Incremental learning is drawing attention to widen capabilities of device-AI. Previous works have researched to reduce numerous computations and memory accesses required for the training process of IL, but they could not show a noticeable improvement in the weight gradient computation (WGC) phase. Therefore, a selective weight update technique that searches for critical weights to be updated is pro-posed by applying the IL algorithm that training per-task binary masks. Also, a novel dataflow for the implementation of selective WGC is introduced on typical NPUs with minimum overheads. On average, the proposed system shows a 2.91x speed up and 2.48x energy efficiency in WGC without degrading training quality.