Research map / Label-efficient and robust learning / Knowledge distillation
DisTaC: Conditioning Task Vectors via Distillation for Robust Model Merging
Kotaro Yoshida, Yuji Naraki, Takafumi Horie, Ryotaro Shimizu, Ioannis Mitliagkas, Hiroki Naganuma
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 |
|---|---|---|---|
| apply DisTaC (norm and confidence conditioning via distillation) before merging | EMR-Merging · 8 vision tasks, Low Confidence | absolute accuracy: 27.4 -> 92.3 (ViT-L-14) | Improved |
| apply DisTaC before merging | TSVM · 8 vision tasks, Low Confidence | absolute accuracy: 60.7 -> 81.5 (ViT-B-32) | Improved |
| apply DisTaC before merging | WUDI-Merging · 8 vision tasks, Norm Mismatch | absolute accuracy: 49.2 -> 84.4 (ViT-B-32) | Improved |
| apply DisTaC before merging | Task arithmetic · 8 vision tasks, Norm Mismatch | absolute accuracy: 63.6 -> 70.0 (ViT-B-32) | Improved |
| apply DisTaC before merging under low confidence | Iso-CTS · 8 vision tasks, Low Confidence | absolute accuracy: 72.5 -> 69.0 (ViT-B-32); 80.8 -> 86.1 (ViT-L-14) | Mixed |
| apply DisTaC under norm mismatch for Iso-CTS ViT-L-14 | Iso-CTS · 8 vision tasks, Norm Mismatch | absolute accuracy: 90.8 -> 90.1 | No clear effect |
| train one task with learning rate 10^-4 instead of 10^-5 (norm disparity) | model merging methods · 8 vision tasks, CLIP ViT-B-32 | normalized accuracy: up to 14% drop | Hurt |
| fine-tune source models with label smoothing 0.1 | model merging methods · 8 vision tasks, CLIP ViT-B-32 | normalized accuracy: up to 24% drop | Hurt |
| increase label smoothing strength during source fine-tuning (ablation) | model merging · 8 vision tasks, CLIP ViT-B-32 | normalized accuracy | Hurt |
| scale task vector down (kappa < 1) (ablation) | task vector · 8 vision tasks, CLIP ViT-B-32 | test accuracy | No clear effect |
| stretch task vector up to kappa=3.0 (ablation) | task vector · 8 vision tasks, CLIP ViT-B-32 | test accuracy | Hurt |
| distill with student temperature 10 and teacher temperature 1 to raise confidence (ablation) | source models · 8 vision tasks, CLIP ViT-B-32 | prediction entropy | Improved |
Closest papers by meaning
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 2024 · similarity 0.89)
- Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent (ICML 2025 · similarity 0.89)
- Whoever Started the interference Should End It: Guiding Data-Free Model Merging via Task Vectors (ICML 2025 · similarity 0.88)
- Twin-Merging: Dynamic Integration of Modular Expertise in Model Merging (NeurIPS 2024 · similarity 0.88)
- CALM: Consensus-Aware Localized Merging for Multi-Task Learning (ICML 2025 · similarity 0.88)
- Scalable Model Merging with Progressive Layer-wise Distillation (ICML 2025 · 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.
