Research map / Learning theory and algorithms / Combinatorial and search optimization
Synthesizing Tasks for Block-based Programming
Umair Ahmed, Maria Christakis, Aleksandr Efremov, Nigel Fernandez, Ahana Ghosh, Abhik Roychoudhury, Adish Singla
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 |
|---|---|---|---|
| Ours vs direct grid mutation (TaskMut) | task synthesis · HOC user study | mean task-solving success: 0.89 vs 0.68 | Improved |
| Ours vs expert-designed Tutor tasks | task synthesis · HOC user study | mean task-solving success: 0.89 vs 0.90 | No clear effect |
| Ours vs TaskMut | task synthesis · HOC user study | mean visual dissimilarity: 2.63 vs 2.17 | Improved |
| Ours vs Tutor | task synthesis · HOC user study | mean visual dissimilarity: 2.63 vs 2.90 | Hurt |
| Ours vs Tutor excluding H2 | task synthesis · HOC user study | mean visual dissimilarity: 2.72 vs 2.93 | No clear effect |
| Ours vs Tutor on nested-conditional tasks H5,H6 | task synthesis · HOC user study | mean visual dissimilarity: 2.94 vs 3.03 | No clear effect |
| Ours vs TaskMut on H5,H6 | task synthesis · HOC user study | mean visual dissimilarity: 1.97 -> 2.94 | Improved |
| Use Z3 jointly with all constraints instead of enumeration plus post-checking (ablation) | mutation stage · HOC/Karel tasks | run-time: 10 to 100 | Improved |
| Use MCTS instead of random search in symbolic execution (ablation) | symbolic execution · H5 task | unique symbolic executions needed: at least 10 times | Improved |
| Increase MCTS iterations (200, 20K, 2M) (ablation) | MCTS · H5 task | F score and generated task quality | Improved |
Closest papers by meaning
- Benchmarking Generative Models on Computational Thinking Tests in Elementary Visual Programming (NeurIPS 2024 · similarity 0.84)
- JanusCoder: Towards a Foundational Visual-Programmatic Interface for Code Intelligence (ICLR 2026 · similarity 0.84)
- Breaking the SFT Plateau: Multimodal Structured Reinforcement Learning for Chart-to-Code Generation (ICLR 2026 · similarity 0.83)
- GenVP: Generating Visual Puzzles with Contrastive Hierarchical VAEs (ICLR 2025 · similarity 0.83)
- CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning (CVPR 2026 · similarity 0.83)
- PRISM: A Benchmark for Programmatic Spatial-Temporal Reasoning (NeurIPS 2026 · similarity 0.83)
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.
