Knowledge Distillation in Deep Learning and its Applications
This is an incremental survey paper that addresses the problem of deploying large models on resource-limited devices for researchers and practitioners.
The paper surveys knowledge distillation techniques for deep learning, proposing a new distillation metric to compare algorithms based on model size and accuracy, and presents conclusions from the analysis.
Deep learning based models are relatively large, and it is hard to deploy such models on resource-limited devices such as mobile phones and embedded devices. One possible solution is knowledge distillation whereby a smaller model (student model) is trained by utilizing the information from a larger model (teacher model). In this paper, we present a survey of knowledge distillation techniques applied to deep learning models. To compare the performances of different techniques, we propose a new metric called distillation metric. Distillation metric compares different knowledge distillation algorithms based on sizes and accuracy scores. Based on the survey, some interesting conclusions are drawn and presented in this paper.