Research map / Architectures and efficiency / Pruning and efficient tuning
FAAR: Efficient Frequency-Aware Multi-Task Fine-Tuning via Automatic Rank Selection
Maxime Fontana, Michael Spratling, Miaojing Shi
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 Performance-Driven Rank Shrinking with DoRA to MTLoRA-style high-rank adaptation (ablation) | MTLoRA (high rank) · PASCAL-Context | Delta m (%): +1.36 -> +4.92 | Improved |
| add Task-Spectral Pyramidal Decoder (frequency filter) (ablation) | FAAR high rank · PASCAL-Context | Delta m (%): +3.83 with TS-PD | Improved |
| add XT-Cons to TS-PD (ablation) | TS-PD · PASCAL-Context | Delta m (%): +3.83 -> +4.63 | Improved |
| combine PDRS with DoRA and TS-PD + XT-Cons (full FAAR) (ablation) | FAAR · PASCAL-Context | Delta m (%): +4.92 -> +5.28 | Improved |
| swap LoRA for DoRA in high-rank MTLoRA (ablation) | MTLoRA (r=64) · PASCAL-Context | Delta m (%): +2.55 -> +1.36 | Hurt |
| swap LoRA for DoRA in low-rank MTLoRA (ablation) | MTLoRA (low rank) · PASCAL-Context | Delta m (%): -2.56 -> -2.31 | Improved |
| use DoRA instead of LoRA together with PDRS (ablation) | high + PDRS · PASCAL-Context | Delta m (%): +2.83 -> +4.92 (TS-PD removed) | Improved |
| add rank shrinking (PDRS) to high-rank MTLoRA with DoRA (ablation) | high + TS-PD + XT-Cons · PASCAL-Context | parameters: rank from 64 to about 5; 3.38M params vs 8.34M MTLoRA | Improved |
| vary initial rank rinit among 64, 32, 16 (ablation) | FAAR · PASCAL-Context | overall performance | No clear effect |
| apply FAAR instead of MTL full fine-tuning | Swin-Tiny MTL · NYUD-v2 | Delta m (%): -8.49 -> -7.88; 28.10M -> 2.85M params | Improved |
| FAAR vs single-task training on small dataset | FAAR · NYUD-v2 | Delta m (%): -7.88 vs 0 single-task | Hurt |
| apply FAAR vs TADFormer | Swin-Tiny MTL PEFT · PASCAL-Context | Delta m (%): +4.24 -> +5.28 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiTASK — DiTASK: Multi-Task Fine-Tuning with Diffeomorphic Transformations (CVPR 2025) · better on 1 of 1 shared result
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
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.87)
- Tuning Stable Rank Shrinkage: Aiming at the Overlooked Structural Risk in Fine-tuning (CVPR 2024 · similarity 0.86)
- PACE: Marrying generalization in PArameter-efficient fine-tuning with Consistency rEgularization (NeurIPS 2024 · similarity 0.86)
- Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation (CVPR 2025 · similarity 0.85)
- MARS: Harmonizing Multimodal Convergence via Adaptive Rank Search (NeurIPS 2026 · similarity 0.85)
- F-Adapter: Frequency-Adaptive Parameter-Efficient Fine-Tuning in Scientific Machine Learning (NeurIPS 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.
