Interactive Program Induction: leaderboard
Models infer hidden programs through examples and queries. Accuracy is reported separately for each interaction and query strategy, preserving the paper’s model identities and task populations.
Source: arxiv.org.
Benchmark results
Results are published separately for each benchmark or evaluation setting.
- Interactive Program Induction - 100 Problems Selection - EIG, unif.
- Interactive Program Induction - 100 Problems Selection - Minimax, unif.
- Interactive Program Induction - 100 Problems Selection - Ours
- Interactive Program Induction - 100 Problems Temperature1 - Approximate
- Interactive Program Induction - 100 Problems Temperature1 - Exact
- Interactive Program Induction - 100 Problems Temperature1 - Uniform
- Interactive Program Induction - Algorithm 1 with EIG (ours) - 1 Seed
- Interactive Program Induction - Algorithm 1 with EIG (ours) - 10 Seed
- Interactive Program Induction - Compute-Matched Propose Verify - Compute-Matched Propose Verify
- Interactive Program Induction - LM only (CodeARC) - 1 Seed
- Interactive Program Induction - LM only (CodeARC) - 10 Seed
- Interactive Program Induction - Query Strategy - Diversity
- Interactive Program Induction - Query Strategy - Prior majority (no query)
- Interactive Program Induction - Query Strategy - Random
- Interactive Program Induction - Query Strategy - Uncertainty
- Interactive Program Induction - Rebuilt From One Seed - Rebuilt From One Seed
Interactive version: theaggregate.ai/benchmark?slug=interactive-program-induction · How It Works · Data refreshed daily, snapshot 2026-11-17.