Research map / Label-efficient and robust learning / Knowledge distillation
Robust Trajectory Distillation: Hybrid Reweighting Meets Teacher-Inspired Targets
KaifengChen KaifengChen, Lechao Cheng, Jiyang Li, Shengeng Tang, FanZhang FanZhang, Yantao Pan, Yaxiong Wang, Tianrui Hui, Zhun Zhong
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 SGR to DATM (ablation) | DATM · CIFAR-10 (40% symmetric noise, IPC=10) | accuracy: 58.2 -> 62.9 | Improved |
| add SGR to DATM (ablation) | DATM · CIFAR-10 (20% symmetric noise, IPC=1000) | accuracy: 81.9 -> 83.3 | Improved |
| add SGR to DATM (ablation) | DATM · CIFAR-10 (40% asymmetric noise, IPC=1000) | accuracy: 71.4 -> 73.3 | Improved |
| add TIAT on top of SGR (ablation) | DATM+SGR · CIFAR-10 (40% symmetric noise, IPC=500) | accuracy: 79.4 -> 80.0 | Improved |
| add TIAT on top of SGR (ablation) | DATM+SGR · CIFAR-10 (40% asymmetric noise, IPC=1000) | accuracy: 73.3 -> 75.1 | Improved |
| add TIAT on top of SGR (ablation) | DATM+SGR · CIFAR-10 (20% symmetric noise, IPC=1000) | accuracy: 83.3 -> 83.3 | No clear effect |
| train teachers with diverse alpha_max instead of fixed alpha_max (ablation) | SGR · CIFAR-10 (symmetric noise) | accuracy | Improved |
| set auxiliary weight beta to 1.0 instead of 0.1 (ablation) | TIAT · CIFAR-10 (20% symmetric noise, IPC=50, 50% subset) | accuracy: 73.3 -> 67.4 | Hurt |
| use beta 0.5 instead of 0.1 (ablation) | TIAT · CIFAR-10 (symmetric noise, IPC=50) | accuracy: 73.5 -> 73.4 | No clear effect |
| full method vs DANCE at IPC=10 | Ours · CIFAR-10 (20% symmetric noise, IPC=10) | accuracy: 68.7 -> 64.5 | Hurt |
| full method vs DATM under high noise | Ours · CIFAR-10 (40% symmetric noise, IPC=500) | accuracy: 74.3 -> 80.0 | Improved |
| full method vs DATM | Ours · CIFAR-100 (40% symmetric noise, IPC=100) | accuracy: 44.4 -> 52.8 | Improved |
Closest papers by meaning
- Distilling What Matters: Confidence-Aware Selective Distillation for Large Language Models (NeurIPS 2026 · similarity 0.86)
- Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling (AAAI 2025 · similarity 0.86)
- Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation (NeurIPS 2026 · similarity 0.86)
- Towards Lossless Dataset Distillation via Difficulty-Aligned Trajectory Matching (ICLR 2024 · similarity 0.86)
- LARK: Learnability-Grounded Trajectory Selection for Efficient Reasoning Distillation (NeurIPS 2026 · similarity 0.86)
- Optimal Transport Reweighting for Robust Learning under Spurious Correlations and Label Noise (NeurIPS 2026 · similarity 0.86)
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.
