Research map / Large language models / Agents, code and math
ShinkaEvolve: Towards Open-Ended and Sample-Efficient Program Evolution
Robert Tjarko Lange, Yuki Imajuku, Edoardo Cetin
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 weighted parent sampling instead of hill climbing or best-of-N (ablation) | ShinkaEvolve · circle packing, AIME, ALE-Bench, MoE | search performance | Improved |
| use bandit-based LLM ensemble selection instead of a single LLM or fixed ensemble (ablation) | ShinkaEvolve · ablation tasks | search performance | Improved |
| add embedding-based code novelty rejection sampling (ablation) | ShinkaEvolve · ablation tasks | search performance | Improved |
| add LLM-as-a-novelty-judge on top of embedding filtering (ablation) | ShinkaEvolve · ablation tasks | search performance | No clear effect |
| evolve ALE-Agent solution with ShinkaEvolve | ALE-Agent · ALE-Bench LITE | average score: approximately 2.3% | Improved |
| evolved MoE load balancing loss with extra under-used-expert regularizer | global-batch LBL (2.7B MoE) · fineweb perplexity and 7 downstream tasks | perplexity / accuracy | Improved |
| evaluate top-5 public solutions on private set (ablation) | ShinkaEvolve ALE-Bench · ALE-Bench LITE | average score: 1923.5 -> 1927.0 | No clear effect |
Closest papers by meaning
- ThetaEvolve: Test-time Learning on Open Problems (ICML 2026 · similarity 0.90)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.89)
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.89)
- DiscoveryBench: Towards Data-Driven Discovery with Large Language Models (ICLR 2025 · similarity 0.88)
- Simple Baselines are Competitive with Code Evolution (NeurIPS 2026 · similarity 0.87)
- REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving (NeurIPS 2025 · similarity 0.87)
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.
