Residual error based knowledge distillation

Knowledge distillation (KD) is one of the most popular ways for model compression. The key idea is to transfer the knowledge from a deep teacher model (T) to a shallower student (S). However, existing methods suffer from performance degradation due to the substantial gap between the learning capacities of S and T. To remedy this problem, this paper proposes Residual error based Knowledge Distillation (RKD), which further distills the knowledge by introducing an assistant model(A). Specifically, S is trained to mimic the feature maps of T, and A aids this process by learning the residual error between them. In this way, S and A complement with each other to get better knowledge from T. Furthermore, we devise an effective method to derive S and A from a given model without increasing the total computational cost. Extensive experiments show that our approach achieves appealing results on popular classification data sets, CIFAR-100 and ImageNet, surpassing state-of-the-art methods and keep strong robustness to adversarial samples. CO 2020 Published by Elsevier B.V.

Automated summary

Knowledge distillation is a popular method for model compression, transferring knowledge from a deep teacher model to a shallower student model. The RKD method introduces an assistant model to further distill knowledge, achieving appealing results on popular classification datasets.