Contrastive Model Inversion for Data-Free Knowledge Distillation

TL;DR

This paper introduces Contrastive Model Inversion (CMI), a novel method that enhances data diversity in model inversion to improve data-free knowledge distillation, outperforming existing techniques in visual plausibility and downstream task effectiveness.

Contribution

The paper proposes a contrastive learning-based approach to explicitly model data diversity in model inversion, addressing mode collapse and improving knowledge distillation performance.

Findings

CMI generates more visually plausible instances than state-of-the-art methods.

CMI achieves significantly better knowledge distillation results.

Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet validate effectiveness.

Abstract

Model inversion, whose goal is to recover training data from a pre-trained model, has been recently proved feasible. However, existing inversion methods usually suffer from the mode collapse problem, where the synthesized instances are highly similar to each other and thus show limited effectiveness for downstream tasks, such as knowledge distillation. In this paper, we propose Contrastive Model Inversion~(CMI), where the data diversity is explicitly modeled as an optimizable objective, to alleviate the mode collapse issue. Our main observation is that, under the constraint of the same amount of data, higher data diversity usually indicates stronger instance discrimination. To this end, we introduce in CMI a contrastive learning objective that encourages the synthesizing instances to be distinguishable from the already synthesized ones in previous batches. Experiments of pre-trained models on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that CMI not only generates more visually plausible instances than the state of the arts, but also achieves significantly superior performance when the generated data are used for knowledge distillation. Code is available at \url{https://github.com/zju-vipa/DataFree}.