Research map / Learning theory and algorithms / Prediction and decision losses
Efficient Adaptive Testing via Gradient Path Matching Subset Selection for AI Education
Yan Zhuang, Junhao Yu, Bohou Zhang, Zachary Pardos, Jinze Wu, Daoqiang Zhang
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 |
|---|---|---|---|
| GPM gradient path matching instead of UATS meta-learning | IRT adaptive testing · EXAM (step 20) | AUC (%): 75.20 -> 77.89 | Improved |
| GPM instead of BECAT global gradient approximation | IRT adaptive testing · ASSIST (step 20) | AUC (%): 73.61 -> 75.27 | Improved |
| GPM instead of NCAT reinforcement learning | IRT adaptive testing · NIPS-EDU (step 20) | AUC (%): 76.66 -> 78.10 | Improved |
| GPM instead of FSI heuristic | IRT adaptive testing · Ability estimation simulation | test length needed: up to 20% fewer questions; 9% on average | Improved |
| reduce training data to 50% for GPM (ablation) | GPM · EXAM | AUC (%): 77.89 -> 74.21 | Hurt |
| reduce training data to 50% for BOBCAT (ablation) | BOBCAT · EXAM | AUC (%): 72.88 -> 70.65 | Hurt |
| add 15% label noise to training for GPM (ablation) | GPM · EXAM | AUC (%): 77.89 -> 73.30 | Hurt |
| add 15% label noise to training for UATS (ablation) | UATS · EXAM | AUC (%): 73.45 -> 71.43 | Hurt |
| GPM under 15% noise vs best baseline GMOCAT | GPM · EXAM 15% noise | AUC (%): 72.50 -> 73.30 | Improved |
Closest papers by meaning
- Explicit and Implicit Examinee-Question Relation Exploiting for Efficient Computerized Adaptive Testing (AAAI 2025 · similarity 0.86)
- A Unified Adaptive Testing System Enabled by Hierarchical Structure Search (ICML 2024 · similarity 0.86)
- Computerized Adaptive Testing via Collaborative Ranking (NeurIPS 2024 · similarity 0.84)
- PT-MoE: An Efficient Finetuning Framework for Integrating Mixture-of-Experts into Prompt Tuning (NeurIPS 2025 · similarity 0.84)
- CurES: From Gradient Analysis to Efficient Curriculum Learning for Reasoning LLMs (ICLR 2026 · similarity 0.84)
- Learning to Discover at Test Time (ICML 2026 · similarity 0.84)
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.
