Research map / Label-efficient and robust learning / Recognition and long-tail
TaskExpert: Dynamically Assembling Multi-Task Representations with Memorial Mixture-of-Experts
Hanrong Ye, Dan Xu
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 MoE (experts and task gating) to the multi-task baseline (ablation) | TaskExpert (ViT-B) · PASCAL-Context | Semseg mIoU: 76.40 -> 77.62 (+1.22) | Improved |
| add context-aware gating to MoE (ablation) | TaskExpert (ViT-B) · PASCAL-Context | Boundary odsF: 70.40 -> 71.70 (+1.30) | Improved |
| add multi-task feature memory (ablation) | TaskExpert (ViT-B) · PASCAL-Context | MTL Perf Delta_m: -2.07 -> -1.46 | Improved |
| use dense instead of sparse (top-1, top-2) expert activation (ablation) | TaskExpert · PASCAL-Context | multi-task performance | Improved |
| use 3x3 instead of 1x1 gating convolution kernel (ablation) | TaskExpert gating · PASCAL-Context | Semseg mIoU: 77.73 -> 78.45 (+0.72) | Improved |
| increase gating kernel from 3x3 to 5x5 (ablation) | TaskExpert gating · PASCAL-Context | Semseg mIoU: 78.45 -> 78.40 | No clear effect |
| replace MLP gating with context-aware gating (ablation) | TaskExpert · PASCAL-Context | MTL Perf Delta_m: -2.96 -> -2.07 | Improved |
| use 12 MMoE modules instead of 3 or 6 (ablation) | TaskExpert (ViT-B) · PASCAL-Context | Semseg mIoU: 78.04 -> 78.45 | Improved |
| TaskExpert versus InvPT with ViT-L backbone | TaskExpert · PASCAL-Context | Semseg mIoU: 79.03 -> 80.64 | Improved |
| TaskExpert versus InvPT with ViT-L backbone | TaskExpert · NYUD-v2 | Semseg mIoU: 53.56 -> 55.35 | Improved |
| TaskExpert ViT-S versus Mod-Squad | TaskExpert ViT-S · PASCAL-Context (ViT-S) | multi-metric: Semseg 74.10 -> 75.04; Boundary 72.00 -> 68.80 | Mixed |
Closest papers by meaning
- TaskPrompter: Spatial-Channel Multi-Task Prompting for Dense Scene Understanding (ICLR 2023 · similarity 0.88)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.88)
- Multi-Task Dense Prediction via Mixture of Low-Rank Experts (CVPR 2024 · similarity 0.88)
- Swiss Army Knife: Synergizing Biases in Knowledge from Vision Foundation Models for Multi-Task Learning (ICLR 2025 · similarity 0.87)
- Towards Minimizing Feature Drift in Model Merging: Layer-wise Task Vector Fusion for Adaptive Knowledge Integration (NeurIPS 2025 · similarity 0.86)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.85)
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.
