Explicit Connection Distillation

1 Jan 2021  ·  Lujun Li, Yikai Wang, Anbang Yao, Yi Qian, Xiao Zhou, Ke He ·

One effective way to ease the deployment of deep neural networks on resource constrained devices is Knowledge Distillation (KD), which boosts the accuracy of a low-capacity student model by mimicking the learnt information of a high-capacity teacher (either a single model or a multi-model ensemble). Although great progress has been attained on KD research, existing efforts are primarily invested to design better distillation losses by using soft logits or intermediate feature representations of the teacher as the extra supervision. In this paper, we present Explicit Connection Distillation (ECD), a new KD framework, which addresses the knowledge distillation problem in a novel perspective of bridging dense intermediate feature connections between a student network and its corresponding teacher generated automatically in the training, achieving knowledge transfer goal via direct cross-network layer-to-layer gradients propagation, without need to define complex distillation losses and assume a pre-trained teacher model to be available. ECD has two interdependent modules. In the first module, given a student network, an auxiliary teacher architecture is temporally generated conditioned on strengthening feature representations of basic convolutions of the student network via replacing them with dynamic additive convolutions and keeping the other layers unchanged in structure. The teacher generated in this way guarantees its superior capacity and makes a perfect feature alignment (both in input and output dimensions) to the student at every convolutional layer. In the second module, dense feature connections between the aligned convolutional layers from the student to its auxiliary teacher are introduced, which allows explicit layer-to-layer gradients propagation from the teacher to the student via the merged model training from scratch. Intriguingly, as feature connection direction is one-way, all feature connections together with the auxiliary teacher merely exist during training phase. Compared to existing KD methods, ECD is more friendly to end users thanks to its simplicity. Experiments on popular image classification tasks validate the effectiveness of our method. Code will be made publicly available.

PDF Abstract


Results from the Paper

  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.