Conditional Pseudo-Supervised Contrast for Data-Free Knowledge Distillation

Data-free knowledge distillation (DFKD) is an effective manner to solve model compression and transmission restrictions while retaining privacy protection, which has attracted extensive attention in recent years. Currently, the majority of existing methods utilize a generator to synthesize images to support the distillation. Although the current methods have achieved great success, there are still many issues to be explored. Firstly, the outstanding performance of supervised learning in deep learning drives us to explore a pseudo-supervised paradigm on DFKD. Secondly, current synthesized methods cannot distinguish the distributions of different categories of samples, thus producing ambiguous samples that may lead to an incorrect evaluation by the teacher. Besides, current methods cannot optimize the category-wise diversity samples, which will hinder the student model learning from diverse samples and further achieving better performance. In this paper, to address the above limitations, we propose a novel learning paradigm, i.e., conditional pseudo-supervised contrast for data-free knowledge distillation (CPSC-DFKD). The primary innovations of CPSC-DFKD are: (1) introducing a conditional generative adversarial network to synthesize category-specific diverse images for pseudo-supervised learning, (2) improving the modules of the generator to distinguish the distributions of different categories, and (3) proposing pseudo-supervised contrastive learning based on teacher and student views to enhance diversity. Comprehensive experiments on three commonly-used datasets validate the performance lift of both the student and generator brought by CPSC-DFKD. The code is available at https://github.com/RoryShao/CPSC-DFKD.git Keywords: model compression, knowledge distillation, representation learning, contrastive learning, privacy protection1. Introduction With the development of artificial intelligence, the deep con-volutional neural networks (DCNNs) have been widely applied in various computer vision tasks and achieved remarkable success, such as image classification [1], object detection [2], and semantic segmentation [3]. Nevertheless, in practical applications, DCNNs suffer from some heavy issues. Firstly, DCNNs always require heavy computation and storage. For example, only to handle one image, a VGG network commonly requires more than 500MB of memory, which makes them hard to be deployed on resource-constrained embedded or edge devices such as mobile phones and autonomous cars.