# Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-level Sparsity via Mixture-of-Experts

arXiv: https://arxiv.org/abs/2305.18691 · HTML: https://mlsanity.com/paper/2305.18691/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- Placed in Vision mixture-of-experts, one of 12 directions in Mixture-of-experts.
- Builds on 26 of its 37 references that are in the map.
- First posted on arXiv 30 May 2023.

## Placed in

- Mixture-of-experts / [Vision mixture-of-experts](https://mlsanity.com/moe/vision-mixture-of-experts/)

## Builds on (most cited first, 26 of 26)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://mlsanity.com/paper/2103.14030/) (2021)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [Training data-efficient image transformers & distillation through attention](https://mlsanity.com/paper/2012.12877/) (2020)
- [The Cityscapes Dataset for Semantic Urban Scene Understanding](https://mlsanity.com/paper/1604.01685/) (2016)
- [Gaussian Error Linear Units (GELUs)](https://mlsanity.com/paper/1606.08415/) (2016)
- [ViViT: A Video Vision Transformer](https://mlsanity.com/paper/2103.15691/) (2021)
- [MLP-Mixer: An all-MLP Architecture for Vision](https://mlsanity.com/paper/2105.01601/) (2021)
- [Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet](https://mlsanity.com/paper/2101.11986/) (2021)
- [Pre-Trained Image Processing Transformer](https://mlsanity.com/paper/2012.00364/) (2020)
- [Taskonomy: Disentangling Task Transfer Learning](https://mlsanity.com/paper/1804.08328/) (2018)
- [Transformer in Transformer](https://mlsanity.com/paper/2103.00112/) (2021)
- [End-to-End Multi-Task Learning with Attention](https://mlsanity.com/paper/1803.10704/) (2018)
- [Cross-stitch Networks for Multi-task Learning](https://mlsanity.com/paper/1604.03539/) (2016)
- [PAD-Net: Multi-Tasks Guided Prediction-and-Distillation Network for Simultaneous Depth Estimation and Scene Parsing](https://mlsanity.com/paper/1805.04409/) (2018)
- [MTI-Net: Multi-Scale Task Interaction Networks for Multi-Task Learning](https://mlsanity.com/paper/2001.06902/) (2020)
- [Pattern-Affinitive Propagation across Depth, Surface Normal and Semantic Segmentation](https://mlsanity.com/paper/1906.03525/) (2019)
- [NDDR-CNN: Layerwise Feature Fusing in Multi-Task CNNs by Neural Discriminative Dimensionality Reduction](https://mlsanity.com/paper/1801.08297/) (2018)
- [FastMoE: A Fast Mixture-of-Expert Training System](https://mlsanity.com/paper/2103.13262/) (2021)
- [M3ViT: Mixture-of-Experts Vision Transformer for Efficient Multi-task Learning with Model-Accelerator Co-design](https://mlsanity.com/paper/2210.14793/) (2022)
- [FTRANS: Energy-Efficient Acceleration of Transformers using FPGA](https://mlsanity.com/paper/2007.08563/) (2020)
- [Hardware Accelerator for Multi-Head Attention and Position-Wise Feed-Forward in the Transformer](https://mlsanity.com/paper/2009.08605/) (2020)
- [Auto-ViT-Acc: An FPGA-Aware Automatic Acceleration Framework for Vision Transformer with Mixed-Scheme Quantization](https://mlsanity.com/paper/2208.05163/) (2022)
- [VAQF: Fully Automatic Software-Hardware Co-Design Framework for Low-Bit Vision Transformer](https://mlsanity.com/paper/2201.06618/) (2022)
- [Vision Transformer using Low-level Chest X-ray Feature Corpus for COVID-19 Diagnosis and Severity Quantification](https://mlsanity.com/paper/2104.07235/) (2021)

## Built on by (most cited first, 23 of 23)

- [EdgeMoE: Empowering Sparse Large Language Models on Mobile Devices](https://mlsanity.com/paper/2308.14352/) (2023)
- [A Survey on Inference Optimization Techniques for Mixture of Experts Models](https://mlsanity.com/paper/2412.14219/) (2024)
- [SiDA-MoE: Sparsity-Inspired Data-Aware Serving for Efficient and Scalable Large Mixture-of-Experts Models](https://mlsanity.com/paper/2310.18859/) (2023)
- [A Survey of Small Language Models](https://mlsanity.com/paper/2410.20011/) (2024)
- [Read-ME: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design](https://mlsanity.com/paper/2410.19123/) (2024)
- [Mixture of Experts (MoE): A Big Data Perspective](https://mlsanity.com/paper/2501.16352/) (2025)
- [Drift-aware Collaborative Assistance Mixture of Experts for Heterogeneous Multistream Learning](https://mlsanity.com/paper/2508.01598/) (2025)
- [TeLLMe: An Energy-Efficient Ternary LLM Accelerator for Prefilling and Decoding on Edge FPGAs](https://mlsanity.com/paper/2504.16266/) (2025)
- [HLSFactory: A Framework Empowering High-Level Synthesis Datasets for Machine Learning and Beyond](https://mlsanity.com/paper/2405.00820/) (2024)
- [HG-PIPE: Vision Transformer Acceleration with Hybrid-Grained Pipeline](https://mlsanity.com/paper/2407.17879/) (2024)
- [UbiMoE: A Ubiquitous Mixture-of-Experts Vision Transformer Accelerator With Hybrid Computation Pattern on FPGA](https://mlsanity.com/paper/2502.05602/) (2025)
- [Understanding and Harnessing Sparsity in Unified Multimodal Models](https://mlsanity.com/paper/2512.02351/) (2025)
- [SMoE: An Algorithm-System Co-Design for Pushing MoE to the Edge via Expert Substitution](https://mlsanity.com/paper/2508.18983/) (2025)
- [When Foresight Pruning Meets Zeroth-Order Optimization: Efficient Federated Learning for Low-Memory Devices](https://mlsanity.com/paper/2405.04765/) (2024)
- [TeLLMe v2: An Efficient End-to-End Ternary LLM Prefill and Decode Accelerator with Table-Lookup Matmul on Edge FPGAs](https://mlsanity.com/paper/2510.15926/) (2025)
- [dMoE: dLLMs with Learnable Block Experts](https://mlsanity.com/paper/2605.30876/) (2026)
- [InterroGate: Learning to Share, Specialize, and Prune Representations for Multi-task Learning](https://mlsanity.com/paper/2402.16848/) (2024)
- [CoQMoE: Co-Designed Quantization and Computation Orchestration for Mixture-of-Experts Vision Transformer on FPGA](https://mlsanity.com/paper/2506.08496/) (2025)
- [Polymorph: Energy-Efficient Multi-Label Classification for Video Streams on Embedded Devices](https://mlsanity.com/paper/2507.14959/) (2025)
- [ViM-Q: Scalable Algorithm-Hardware Co-Design for Vision Mamba Model Inference on FPGA](https://mlsanity.com/paper/2605.01935/) (2026)
- [ELiTeFormer: An Efficient Transformer for FPGAs](https://mlsanity.com/paper/2607.03652/) (2026)
- [MoTE: Mixture of Task Experts for Multi-Task Video Understanding](https://mlsanity.com/paper/2608.24763/) (2026)
- [Mira: Memory-Efficient MoE Inference Using Adaptive Caching and Predictive Expert Staging](https://mlsanity.com/paper/2609.38090/) (2026)

## Abstract (arXiv)

Computer vision researchers are embracing two promising paradigms: Vision Transformers (ViTs) and Multi-task Learning (MTL), which both show great performance but are computation-intensive, given the quadratic complexity of self-attention in ViT and the need to activate an entire large MTL model for one task. M3ViT is the latest multi-task ViT model that introduces mixture-of-experts (MoE), where only a small portion of subnetworks ("experts") are sparsely and dynamically activated based on the current task. M3ViT achieves better accuracy and over 80% computation reduction but leaves challenges for efficient deployment on FPGA. Our work, dubbed Edge-MoE, solves the challenges to introduce the first end-to-end FPGA accelerator for multi-task ViT with a collection of architectural innovations, including (1) a novel reordering mechanism for self-attention, which requires only constant bandwidth regardless of the target parallelism; (2) a fast single-pass softmax approximation; (3) an accurate and low-cost GELU approximation; (4) a unified and flexible computing unit that is shared by almost all computational layers to maximally reduce resource usage; and (5) uniquely for M3ViT, a novel patch reordering method to eliminate memory access overhead. Edge-MoE achieves 2.24x and 4.90x better energy efficiency comparing with GPU and CPU, respectively. A real-time video demonstration is available online, along with our open-source code written using High-Level Synthesis.