Research map / Probabilistic and scientific modeling / Variational and generative sampling
Sampling for Bayesian Program Learning
Kevin Ellis, Armando Solar-Lezama, Josh Tenenbaum
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 |
|---|---|---|---|
| remove tilt correction by setting d to |x*| - n (ablation) | ProgramSample · Text edit | samples produced in under an hour: 49 -> > 1 hour | Hurt |
| remove tilt correction on counting task (ablation) | ProgramSample · List manipulation: count | tractability | Hurt |
| pure rejection sampling from the prior | baseline · Text edit | feasibility | Hurt |
| sample programs instead of MDL point estimate: held-out accuracy | ProgramSample · Text edit held-out | generalization accuracy | No clear effect |
| offer top C sampled predictions instead of MDL single prediction | ProgramSample · Text edit one-shot | problems solved | Improved |
| vary training set size (small to large) for recursive list routines | ProgramSample · List manipulation: sort | solver time per sample: 463 -> 1549 | Hurt |
| vary training set size (small to large) for reverse | ProgramSample · List manipulation: reverse | solver time per sample: 39 -> 326 | Hurt |
| vary training set size on text edit | ProgramSample · Text edit | solver time per sample: 21 -> 84 | Mixed |
| more training examples (5-11) for list routines | ProgramSample · List manipulation | correct implementation recovered | Improved |
| tilt correction and random XOR constraints vs ground-truth distribution (ablation) | ProgramSample · Counting task, 80 programs | sampling frequency vs ground truth | No clear effect |
| recursive routines vs non-recursive count | ProgramSample · List manipulation | solver time: 1549 -> 1 | Hurt |
Closest papers by meaning
- Constrained Sampling for Language Models Should Be Easy: An MCMC Perspective (NeurIPS 2025 · similarity 0.86)
- Program Synthesis via Test-Time Transduction (NeurIPS 2025 · similarity 0.85)
- Practical Bayesian Algorithm Execution via Posterior Sampling (NeurIPS 2024 · similarity 0.85)
- Reinforcement Learning and Data-Generation for Syntax-Guided Synthesis (AAAI 2024 · similarity 0.84)
- Bayesian Program Learning by Decompiling Amortized Knowledge (ICML 2024 · similarity 0.84)
- Learning to Learn in Interactive Constraint Acquisition (AAAI 2024 · 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.
