Research map / Architectures and efficiency / Pruning and efficient tuning
AdaRank: Adaptive Rank Pruning for Enhanced Model Merging
Chanhyuk Lee, Jiho Choi, Chanryeol Lee, Donggyun Kim, Seunghoon Hong
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 AdaRank to Task Arithmetic | Task Arithmetic · ViT-B/32, 8 tasks | average accuracy: 69.2 -> 87.9 | Improved |
| add AdaRank to Iso-CTS | Iso-CTS · ViT-B/32, 8 tasks | average accuracy: 84.9 -> 89.4 | Improved |
| add AdaRank to TSV-M | TSV-M · ViT-B/32, 20 tasks | average accuracy: 76.7 -> 86.9 | Improved |
| add AdaRank instead of AdaMerging to Task Arithmetic | TA+AdaMerging · ViT-B/32, 8 tasks | average accuracy: 80.1 -> 87.9 | Improved |
| add AdaRank instead of AdaMerging to Iso-CTS | Iso-CTS+AdaMerging · ViT-L/14, 8 tasks | average accuracy: 95.4 -> 95.5 | Improved |
| add AdaRank to Task Arithmetic | Task Arithmetic · GLUE 7 tasks, RoBERTa-base | average performance: 0.6718 -> 0.7032 | Improved |
| add AdaRank to CART | CART · GLUE 7 tasks, GPT-2 | average performance: 0.6182 -> 0.6587 | Improved |
| extend AdaRank mask from top-16% range to full range (ablation) | AdaRank on TA · ViT-B/32, 8 tasks | average accuracy: 87.5 -> 87.9 | Improved |
| restrict to top-16% fixed vs adaptive masking in top-16% range (ablation) | Top-16% Fixed TA · ViT-B/32, 8 tasks | average accuracy: 68.8 -> 87.5 | Improved |
| adapt mask B in addition to lambda (ablation) | TA with lambda only · ViT-B/32, 8 tasks | average accuracy: 80.1 -> 87.9 | Improved |
| adapt lambda in addition to mask B (ablation) | CART with B only · ViT-B/32, 8 tasks | average accuracy: 88.7 -> 89.2 | Improved |
| adapt only mask B instead of only lambda (ablation) | CART · ViT-B/32, 8 tasks | average accuracy: 85.9 -> 88.7 | Improved |
Closest papers by meaning
- Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent (ICML 2025 · similarity 0.90)
- Whoever Started the interference Should End It: Guiding Data-Free Model Merging via Task Vectors (ICML 2025 · similarity 0.90)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 2024 · similarity 0.90)
- Dynamic Model Merging Made Slim (NeurIPS 2026 · similarity 0.89)
- No Task Left Behind: Isotropic Model Merging with Common and Task-Specific Subspaces (ICML 2025 · similarity 0.89)
- Contribution-Aware Structured Sparsity for Model Merging (NeurIPS 2026 · similarity 0.88)
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.
