Scaling Vision with Sparse Mixture of Experts

by   Carlos Riquelme, et al.

Sparsely-gated Mixture of Experts networks (MoEs) have demonstrated excellent scalability in Natural Language Processing. In Computer Vision, however, almost all performant networks are "dense", that is, every input is processed by every parameter. We present a Vision MoE (V-MoE), a sparse version of the Vision Transformer, that is scalable and competitive with the largest dense networks. When applied to image recognition, V-MoE matches the performance of state-of-the-art networks, while requiring as little as half of the compute at inference time. Further, we propose an extension to the routing algorithm that can prioritize subsets of each input across the entire batch, leading to adaptive per-image compute. This allows V-MoE to trade-off performance and compute smoothly at test-time. Finally, we demonstrate the potential of V-MoE to scale vision models, and train a 15B parameter model that attains 90.35 ImageNet.


page 6

page 31

page 32

page 33

page 36

page 41

page 42


An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

While the Transformer architecture has become the de-facto standard for ...

GLaM: Efficient Scaling of Language Models with Mixture-of-Experts

Scaling language models with more data, compute and parameters has drive...

A Review of Sparse Expert Models in Deep Learning

Sparse expert models are a thirty-year old concept re-emerging as a popu...

Biased Mixtures Of Experts: Enabling Computer Vision Inference Under Data Transfer Limitations

We propose a novel mixture-of-experts class to optimize computer vision ...

Changing Model Behavior at Test-Time Using Reinforcement Learning

Machine learning models are often used at test-time subject to constrain...

A Theoretical View on Sparsely Activated Networks

Deep and wide neural networks successfully fit very complex functions to...

Sparse Mixers: Combining MoE and Mixing to build a more efficient BERT

We combine the capacity of sparsely gated Mixture-of-Experts (MoE) with ...