Boosting the Cross-Architecture Generalization of Dataset Distillation through an Empirical Study
This addresses a specific bottleneck in dataset distillation for researchers and practitioners, but it is incremental as it builds on existing methods to improve generalization.
The paper tackles the problem of poor cross-architecture generalization in dataset distillation, which limits its practical use, and proposes a method called ELF that enhances this generalization by using features from intermediate layers for evaluation, resulting in improved performance across different architectures.
The poor cross-architecture generalization of dataset distillation greatly weakens its practical significance. This paper attempts to mitigate this issue through an empirical study, which suggests that the synthetic datasets undergo an inductive bias towards the distillation model. Therefore, the evaluation model is strictly confined to having similar architectures of the distillation model. We propose a novel method of EvaLuation with distillation Feature (ELF), which utilizes features from intermediate layers of the distillation model for the cross-architecture evaluation. In this manner, the evaluation model learns from bias-free knowledge therefore its architecture becomes unfettered while retaining performance. By performing extensive experiments, we successfully prove that ELF can well enhance the cross-architecture generalization of current DD methods. Code of this project is at \url{https://github.com/Lirui-Zhao/ELF}.