Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Giving Feedback on Interactive Student Programs with Meta-Exploration
Evan Zheran Liu, Moritz Pascal Stephan, Allen Nie, Christopher J Piech, Emma Brunskill, Chelsea Finn
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 |
|---|---|---|---|
| use per-error exploration policies instead of one policy for all errors (ablation) | Meta-Exploration · Bounce (8-error rubric) | accuracy: 91.3 ± 0.4% -> 94.3 ± 1.3% | Improved |
| provide shaped per-timestep exploration rewards instead of directly maximizing end-of-episode accuracy (ablation) | Meta-Exploration · Bounce (8-error rubric) | accuracy: 84.8 ± 2.2% -> 94.3 ± 1.3% | Improved |
| use meta-exploration for fine-grained feedback rather than the prior Play-to-Grade approach | Bounce (8-error rubric) · Bounce (8-error rubric) | accuracy: 17.7% greater accuracy | Improved |
| test on held-out ball and paddle speeds | Meta-Exploration · Bounce (held-out ball and paddle speeds) | accuracy: 89.0 ± 1.9%; 89.7 ± 1.7%; 89.6 ± 2.0%; 89.3 ± 1.8% | No clear effect |
| train and evaluate across all 28 rubric errors rather than only 8 | Meta-Exploration · Bounce (28 error types) | accuracy: 94%; 94.3% | No clear effect |
| learn to target the paddle-skewer behavior through program interaction | Meta-Exploration · Breakout (paddle skewer) | accuracy: 93.8% | Improved |
| learn to detect the does-not-delete-brick error through program interaction | Meta-Exploration · Breakout (does not delete brick) | accuracy: 96.9% | Improved |
Closest papers by meaning
- Students’ Perceptions and Preferences of Generative Artificial Intelligence Feedback for Programming (AAAI 2024 · similarity 0.84)
- Hints-In-Browser: Benchmarking Language Models for Programming Feedback Generation (NeurIPS 2024 · similarity 0.81)
- Simulating Students or Sycophantic Problem Solving? On Misconception Faithfulness of LLM Simulators (NeurIPS 2026 · similarity 0.80)
- Counterexample Guided Program Repair Using Zero-Shot Learning and MaxSAT-based Fault Localization (AAAI 2025 · similarity 0.80)
- Benchmarking Generative Models on Computational Thinking Tests in Elementary Visual Programming (NeurIPS 2024 · similarity 0.80)
- BEAGLE: Behavior-Enforced Agent for Grounded Learner Emulation (NeurIPS 2026 · similarity 0.80)
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.
