Research map / Label-efficient and robust learning / Recognition and long-tail
Selective Weak-to-Strong Generalization
Hao Lang, Fei Huang, Yongbin Li
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 P(IK) classifier for selective labels (ablation) | selective W2SG · SciQ (Qwen-1.8B weak) | accuracy: 89.86 -> 90.60 | Improved |
| add P(IK) classifier for selective labels (ablation) | selective W2SG · CosmosQA (Qwen-1.8B weak) | accuracy: 81.36 -> 83.85 | Improved |
| add graph-smoothed weak labels (ablation) | selective W2SG · CosmosQA (Qwen-1.8B weak) | accuracy: 83.12 -> 83.85 | Improved |
| add graph-smoothed weak labels (ablation) | selective W2SG · BoolQ (Qwen-1.8B weak) | accuracy: 83.11 -> 83.39 | Improved |
| use selective labels instead of multi-task P(IK) head (ablation) | selective W2SG · SciQ (Qwen-1.8B weak) | accuracy: 89.98 -> 90.60 | Improved |
| replace naive finetuning with selective W2SG | Qwen-14B student · SciQ (GPT-2 weak) | accuracy: 77.90 -> 83.30 | Improved |
| replace naive finetuning with selective W2SG | Qwen-14B student · BoolQ (GPT-2 weak) | accuracy: 65.89 -> 68.64 | Improved |
| replace auxiliary confidence loss with selective W2SG | Qwen-14B student · BoolQ (GPT-2 weak) | PGR: -10.76 -> 14.92 | Improved |
| replace auxiliary confidence loss with selective W2SG | Qwen-14B student · CosmosQA (Qwen-1.8B weak) | PGR: 20.87 -> 44.57 | Improved |
| add auxiliary confidence loss to naive finetuning | Qwen-14B student · BoolQ (GPT-2 weak) | accuracy: 65.89 -> 62.22 | Hurt |
| train P(IK) on CosmosQA instead of Winograde (ablation) | selective W2SG · SciQ | accuracy: 89.80 -> 90.70 | Improved |
| train P(IK) on ARC instead of CosmosQA (ablation) | selective W2SG · BoolQ | accuracy: 83.20 -> 83.39 | Improved |
Closest papers by meaning
- Strong Empowered and Aligned Weak Mastered Annotation for Weak-to-Strong Generalization (AAAI 2025 · similarity 0.93)
- Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision (ICML 2024 · similarity 0.92)
- Weak-to-Strong Generalization under Distribution Shifts (NeurIPS 2025 · similarity 0.92)
- Debate Helps Weak-to-Strong Generalization (AAAI 2025 · similarity 0.92)
- Representations Shape Weak-to-Strong Generalization: Theoretical Insights and Empirical Predictions (ICML 2025 · similarity 0.91)
- Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization (ICLR 2025 · similarity 0.90)
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.
