Research map / Architectures and efficiency / Pruning and efficient tuning
Frequency Switching Mechanism for Parameter-Efficient Multi-Task Learning
Shih-Wen Liu, Yen-Chang Chen, Wei-Ta Chu, Fu-En Yang, Yu-Chiang Frank Wang
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 sine transformation to shared base (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +3.67 -> +5.39 | Improved |
| add Gaussian low-pass filter (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +4.82 -> +5.39 | Improved |
| add Lightweight Clock Net (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.12 -> +5.39 | Improved |
| share one base matrix across tasks instead of independent bases (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.03 -> +5.39 | Improved |
| share decoder weights via frequency switching instead of independent decoders (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.31 -> +5.39 | Improved |
| increase rank 16 to 64 (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +3.47 -> +5.39 | Improved |
| swap decoder HRNet to ASPP (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.80 -> +7.14 | Improved |
| swap decoder HRNet to SegFormer (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.80 -> +5.86 | No clear effect |
| Gaussian low-pass sigma 1.0 to 0.5 (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.39 -> +4.90 | Hurt |
| Gaussian kernel size 7 to 5 or 9 (ablation) | Free Sinewich · PASCAL-Context | Delta m (%): +5.39 -> +5.32 | No clear effect |
| use ImageNet-22K instead of ImageNet-1K pretraining (ablation) | Free Sinewich · PASCAL-Context | Delta m | Improved |
| Free Sinewich r=64 on NYUDv2 vs single-task | Free Sinewich · NYUDv2 | Delta m (%): -0.52 | Hurt |
Closest papers by meaning
- MTLoRA: Low-Rank Adaptation Approach for Efficient Multi-Task Learning (CVPR 2024 · similarity 0.87)
- NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective (NeurIPS 2025 · similarity 0.85)
- Parameter-Efficient Multi-Task Model Fusion with Partial Linearization (ICLR 2024 · similarity 0.85)
- MmAP: Multi-Modal Alignment Prompt for Cross-Domain Multi-Task Learning (AAAI 2024 · similarity 0.85)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.84)
- Less is More: Efficient Model Merging with Binary Task Switch (CVPR 2025 · similarity 0.84)
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.
