Abstract
Knowledge distillation is the process of transferring the knowledge from a large model to a small model. In this process, the small model learns the generalization ability of the large model and retains the performance close to that of the large model. Knowledge distillation provides a training means to migrate the knowledge of models, facilitating model deployment and speeding up inference. However, previous distillation methods require pre-trained teacher models, which still bring computational and storage overheads. In this paper, a novel general training framework called Self-Feature Regularization~(SFR) is proposed, which uses features in the deep layers to supervise feature learning in the shallow layers, retains more semantic information. Specifically, we firstly use EMD-l2 loss to match local features and a many-to-one approach to distill features more intensively in the channel dimension. Then dynamic label smoothing is used in the output layer to achieve better performance. Experiments further show the effectiveness of our proposed framework.
Abstract (translated)
URL
https://arxiv.org/abs/2103.07350
