Can Attention Enable MLPs To Catch Up With CNNs?

Abstract

In the first week of May, 2021, researchers from four different institutions: Google, Tsinghua University, Oxford University and Facebook, shared their latest work [16,7,12,17] on arXiv.org almost at the same time, each proposing new learning architectures, consisting mainly of linear layers, claiming them to be comparable, or even superior to convolutional-based models. This sparked immediate discussion and debate in both academic and industrial communities as to whether MLPs are sufficient, many thinking that learning architectures are returning to MLPs. Is this true?In this perspective, we give a brief history of learning architectures, including multilayer perceptrons (MLPs), convolutional neural networks (CNNs) and transformers. We then examine what the four newly proposed architectures have in common. Finally, we give our views on challenges and directions for new learning architectures, hoping to inspire future research.

Multi-Exit Vision Transformer for Dynamic Inference

“…Several MLP-based architectures for computer vision that also operate on sequences of image patches have been recently proposed [7]. The aim of these architectures is to reduce the computational cost of ViT by removing the attention mechanism, while achieving a comparable performance by preserving a global receptive field similar to that of ViT.…”