Partial to Whole Knowledge Distillation: Progressive Distilling Decomposed Knowledge Boosts Student Better (2109.12507v1)

Abstract: Knowledge distillation field delicately designs various types of knowledge to shrink the performance gap between compact student and large-scale teacher. These existing distillation approaches simply focus on the improvement of \textit{knowledge quality}, but ignore the significant influence of \textit{knowledge quantity} on the distillation procedure. Opposed to the conventional distillation approaches, which extract knowledge from a fixed teacher computation graph, this paper explores a non-negligible research direction from a novel perspective of \textit{knowledge quantity} to further improve the efficacy of knowledge distillation. We introduce a new concept of knowledge decomposition, and further put forward the \textbf{P}artial to \textbf{W}hole \textbf{K}nowledge \textbf{D}istillation~(\textbf{PWKD}) paradigm. Specifically, we reconstruct teacher into weight-sharing sub-networks with same depth but increasing channel width, and train sub-networks jointly to obtain decomposed knowledge~(sub-networks with more channels represent more knowledge). Then, student extract partial to whole knowledge from the pre-trained teacher within multiple training stages where cyclic learning rate is leveraged to accelerate convergence. Generally, \textbf{PWKD} can be regarded as a plugin to be compatible with existing offline knowledge distillation approaches. To verify the effectiveness of \textbf{PWKD}, we conduct experiments on two benchmark datasets:~CIFAR-100 and ImageNet, and comprehensive evaluation results reveal that \textbf{PWKD} consistently improve existing knowledge distillation approaches without bells and whistles.