Research map / Architectures and efficiency / Mixture of experts
TAG-MoE: Task-Aware Gating for Unified Generative Mixture-of-Experts
Yu Xu, Hongbin Yan, Juan Cao, Yiji Cheng, Tiankai Hang, Runze He, Zijin Yin, Shiyi Zhang, Yuxin Zhang, Jintao Li, Chunyu Wang, Qinglin Lu and 2 more
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add predictive alignment regularization (Lalign) to the MoE gate (ablation) | TAG-MoE · ICE-Bench | vllmqa: 0.677 -> 0.847 | Improved |
| add predictive alignment regularization (Lalign) (ablation) | TAG-MoE · ICE-Bench | Face-ref: 0.3779 -> 0.4642 | Improved |
| add predictive alignment regularization (Lalign) (ablation) | TAG-MoE · ICE-Bench | DINO-ref: 0.7355 -> 0.7620 | Improved |
| replace dense FFN with sparse MoE of equal activated parameters (no Lalign) (ablation) | Qwen-Image based unified model · ICE-Bench | vllmqa: 0.637 -> 0.677 | Improved |
| replace dense FFN with MoE with Lalign (ablation) | Qwen-Image based unified model · ICE-Bench | vllmqa: 0.637 -> 0.847 | Improved |
| replace dense FFN with MoE with Lalign (ablation) | Qwen-Image based unified model · ICE-Bench | Face-ref: 0.3544 -> 0.4642 | Improved |
| use MoE with Lalign vs dense on CLIP-cap (ablation) | Qwen-Image based unified model · ICE-Bench | CLIP-cap: 0.263 -> 0.281 | Improved |
| TAG-MoE vs best open-source editing baseline | TAG-MoE · EmuEdit-bench | vllmqa: 0.9174 -> 0.9284 | Improved |
| TAG-MoE vs best open-source on source fidelity | TAG-MoE · ICE-Bench | CLIP-src: 0.866 (DreamOmni2) vs 0.857 | Hurt |
| TAG-MoE vs specialized subject-driven models | TAG-MoE · DreamBench++ | DINO-ref: 0.7613 vs DreamOmni2 0.8008 | Mixed |
Closest papers by meaning
- Diff-MoE: Diffusion Transformer with Time-Aware and Space-Adaptive Experts (ICML 2025 · similarity 0.87)
- Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE (NeurIPS 2026 · similarity 0.87)
- Learning to Specialize: Joint Gating-Expert Training for Adaptive MoEs in Decentralized Settings (NeurIPS 2025 · similarity 0.86)
- On the Expressive Power of Mixture-of-Experts for Structured Complex Tasks (NeurIPS 2025 · similarity 0.86)
- MoEUT: Mixture-of-Experts Universal Transformers (NeurIPS 2024 · similarity 0.86)
- Mixture of Tokens: Continuous MoE through Cross-Example Aggregation (NeurIPS 2024 · similarity 0.86)
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.
