AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

M³ViT: Mixture-of-Experts Vision Transformer for Efficient Multi-task Learning with Model-Accelerator Co-design

hanxue liang, Zhiwen Fan, Rishov Sarkar, Ziyu Jiang, Tianlong Chen, Kai Zou, Yu Cheng, Cong Hao, Zhangyang Wang

NeurIPS 2022 · Architectures and efficiency › Mixture of experts · approach: Sparse MoE as regularization and attention

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace dense ViT feed-forward layers with task-dependent sparse MoEM-ViT · PASCAL-ContextΔm (%): -1.77 -> +2.71Improved
Replace dense ViT feed-forward layers with task-dependent sparse MoEM-ViT · NYUD-v2Δm (%): −6.27 -> +1.59Improved
Add FPGA computation reordering to MoE inferenceMoE ViT-small on FPGA · PASCAL-ContextEnergy (W·s): 6.375 -> 0.690Improved
Add FPGA computation reordering to MoE inferenceMoE ViT-small on FPGA · NYUD-v2Energy (W·s): 7.506 -> 0.845Improved
Replace a single shared router with separate task-specific routers (ablation)M³ViT · PASCAL-ContextΔm (%): +1.40 -> +2.71Improved
Replace a single shared router with separate task-specific routers (ablation)M³ViT · NYUD-v2Δm (%): +0.31 -> +1.59Improved
Use task-conditioned shared routing instead of separate task routers (ablation)M³ViT · PASCAL-ContextΔm (%): +2.71 -> +2.22Hurt
Use task-conditioned shared routing instead of separate task routers (ablation)M³ViT · NYUD-v2Δm (%): +1.59 -> +0.74Hurt
Remove computation reordering and use cache-based expert loading (ablation)MoE ViT-small on FPGA · PASCAL-ContextEnergy (W·s): 0.690 -> 6.375Hurt
Remove computation reordering and use cache-based expert loading (ablation)MoE ViT-small on FPGA · NYUD-v2Energy (W·s): 0.845 -> 7.506Hurt
Scale the MoE ViT backbone from ViT-small to ViT-baseM³ViT · PASCAL-ContextΔm (%): +2.71 -> +4.00Improved
Scale the MoE ViT backbone from ViT-small to ViT-baseM³ViT · NYUD-v2Δm (%): +1.59 -> +8.32Improved

Later papers that compared with this method

2 later accepted papers report M³ViT as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy