Large-Scale Generative Data-Free Distillation
This work provides a more efficient and scalable solution for data-free knowledge distillation, which is crucial for practitioners facing privacy, proprietary, or availability constraints on original training data.
The paper addresses the challenge of data-free knowledge distillation by proposing a method that trains a generative image model using the intrinsic normalization layer statistics of a pre-trained teacher network. This approach enables the creation of an ensemble of generators to produce substitute inputs for distillation, achieving 95.02% on CIFAR-10 and 77.02% on CIFAR-100, and successfully scaling to ImageNet.
Knowledge distillation is one of the most popular and effective techniques for knowledge transfer, model compression and semi-supervised learning. Most existing distillation approaches require the access to original or augmented training samples. But this can be problematic in practice due to privacy, proprietary and availability concerns. Recent work has put forward some methods to tackle this problem, but they are either highly time-consuming or unable to scale to large datasets. To this end, we propose a new method to train a generative image model by leveraging the intrinsic normalization layers' statistics of the trained teacher network. This enables us to build an ensemble of generators without training data that can efficiently produce substitute inputs for subsequent distillation. The proposed method pushes forward the data-free distillation performance on CIFAR-10 and CIFAR-100 to 95.02% and 77.02% respectively. Furthermore, we are able to scale it to ImageNet dataset, which to the best of our knowledge, has never been done using generative models in a data-free setting.