Are Large Kernels Better Teachers than Transformers for ConvNets?

Tianjin Huang, Lu Yin, Zhenyu Zhang, Li Shen, Meng Fang, Mykola Pechenizkiy, Zhangyang Wang, Shiwei Liu
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:14023-14038, 2023.

Abstract

This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small-kernel ConvNets. While Transformers have led state-of-the-art (SOTA) performance in various fields with ever-larger models and labeled data, small-kernel ConvNets are considered more suitable for resource-limited applications due to the efficient convolution operation and compact weight sharing. KD is widely used to boost the performance of small-kernel ConvNets. However, previous research shows that it is not quite effective to distill knowledge (e.g., global information) from Transformers to small-kernel ConvNets, presumably due to their disparate architectures. We hereby carry out a first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel ConvNets, due to more similar architectures. Our findings are backed up by extensive experiments on both logit-level and feature-level KD "out of the box", with no dedicated architectural nor training recipe modifications. Notably, we obtain the best-ever pure ConvNet under 30M parameters with 83.1% top-1 accuracy on ImageNet, outperforming current SOTA methods including ConvNeXt V2 and Swin V2. We also find that beneficial characteristics of large-kernel ConvNets, e.g., larger effective receptive fields, can be seamlessly transferred to students through this large-to-small kernel distillation. Code is available at: https://github.com/VITA-Group/SLaK.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-huang23o, title = {Are Large Kernels Better Teachers than Transformers for {C}onv{N}ets?}, author = {Huang, Tianjin and Yin, Lu and Zhang, Zhenyu and Shen, Li and Fang, Meng and Pechenizkiy, Mykola and Wang, Zhangyang and Liu, Shiwei}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {14023--14038}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/huang23o/huang23o.pdf}, url = {https://proceedings.mlr.press/v202/huang23o.html}, abstract = {This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small-kernel ConvNets. While Transformers have led state-of-the-art (SOTA) performance in various fields with ever-larger models and labeled data, small-kernel ConvNets are considered more suitable for resource-limited applications due to the efficient convolution operation and compact weight sharing. KD is widely used to boost the performance of small-kernel ConvNets. However, previous research shows that it is not quite effective to distill knowledge (e.g., global information) from Transformers to small-kernel ConvNets, presumably due to their disparate architectures. We hereby carry out a first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel ConvNets, due to more similar architectures. Our findings are backed up by extensive experiments on both logit-level and feature-level KD "out of the box", with no dedicated architectural nor training recipe modifications. Notably, we obtain the best-ever pure ConvNet under 30M parameters with 83.1% top-1 accuracy on ImageNet, outperforming current SOTA methods including ConvNeXt V2 and Swin V2. We also find that beneficial characteristics of large-kernel ConvNets, e.g., larger effective receptive fields, can be seamlessly transferred to students through this large-to-small kernel distillation. Code is available at: https://github.com/VITA-Group/SLaK.} }
Endnote
%0 Conference Paper %T Are Large Kernels Better Teachers than Transformers for ConvNets? %A Tianjin Huang %A Lu Yin %A Zhenyu Zhang %A Li Shen %A Meng Fang %A Mykola Pechenizkiy %A Zhangyang Wang %A Shiwei Liu %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-huang23o %I PMLR %P 14023--14038 %U https://proceedings.mlr.press/v202/huang23o.html %V 202 %X This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small-kernel ConvNets. While Transformers have led state-of-the-art (SOTA) performance in various fields with ever-larger models and labeled data, small-kernel ConvNets are considered more suitable for resource-limited applications due to the efficient convolution operation and compact weight sharing. KD is widely used to boost the performance of small-kernel ConvNets. However, previous research shows that it is not quite effective to distill knowledge (e.g., global information) from Transformers to small-kernel ConvNets, presumably due to their disparate architectures. We hereby carry out a first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel ConvNets, due to more similar architectures. Our findings are backed up by extensive experiments on both logit-level and feature-level KD "out of the box", with no dedicated architectural nor training recipe modifications. Notably, we obtain the best-ever pure ConvNet under 30M parameters with 83.1% top-1 accuracy on ImageNet, outperforming current SOTA methods including ConvNeXt V2 and Swin V2. We also find that beneficial characteristics of large-kernel ConvNets, e.g., larger effective receptive fields, can be seamlessly transferred to students through this large-to-small kernel distillation. Code is available at: https://github.com/VITA-Group/SLaK.
APA
Huang, T., Yin, L., Zhang, Z., Shen, L., Fang, M., Pechenizkiy, M., Wang, Z. & Liu, S.. (2023). Are Large Kernels Better Teachers than Transformers for ConvNets?. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:14023-14038 Available from https://proceedings.mlr.press/v202/huang23o.html.

Related Material