EliteTable: Self-Improving Program Search for Executable Control Policies
Quick Answer
EliteTable is a self-improving program-search algorithm leveraging large language models, demonstrating significant performance gains across six Gymnasium control tasks.
Quick Take
Notably, in CartPole-v1, the selected winner achieved the search-score ceiling in the first generation, while improvements were observed in four of six tasks, though some tasks showed declines in performance.
Key Points
- EliteTable uses a fixed-capacity table of high-scoring solutions for program search.
- Performance evaluated across six Gymnasium tasks with ten generations each.
- In CartPole-v1, the winner reached the search-score ceiling in the first generation.
- Improvements were noted in four out of six tasks, with some declines in others.
- Further comparisons are needed to assess EliteTable's relative performance.
DeepSignal Analysis
What happened
EliteTable is a program-search algorithm that utilizes large language models to enhance control policies. It was tested on six Gymnasium control tasks, showing overall performance improvements in five tasks, with CartPole-v1 achieving peak performance in the first generation. However, some tasks, like Acrobot-v1, experienced declines in performance during certain searches.
Key evidence
- In CartPole-v1, the selected winner reached the search-score ceiling in the first generation across all ten searches.
- Candidate-score medians improved from the first to the final generation in all six Gymnasium tasks.
- Six out of ten searches for Acrobot-v1 showed a decline in performance on held-out episodes, indicating variability in results.
Why it matters
The findings suggest that EliteTable can effectively leverage large language models for program search, potentially leading to advancements in automated control policy generation. However, the mixed results across tasks highlight the need for further investigation into the algorithm's robustness and the specific contributions of its components. Understanding these factors is crucial for assessing its applicability in broader AI contexts.
Paper Resources
📖 Reader Mode
~2 min readAbstract:EliteTable is a self-improving program-search algorithm guided by large language models (LLMs) that uses a fixed-capacity table of high-scoring solutions to guide new proposals, edits of existing solutions, and remixes of several solutions. Performance was evaluated across six Gymnasium control tasks, each with ten generations and fifty candidate slots per generation. Each task was evaluated across ten independently initialized searches. Candidate-score medians rose between the first and final generations in all six tasks, and selected-winner search-score medians rose in five. In CartPole-v1, the selected winner reached the search-score ceiling in the first generation of all ten searches. All ten paired held-out endpoints improved in four of the six tasks. BipedalWalker-v3 and CarRacing-v3 improved without reaching their reporting targets. Six of ten Acrobot-v1 searches declined on held-out episodes, and the median paired change was negative. The results document improvement during program search and the transfer of selected-winner gains to new episodes in some tasks. Comparisons and ablations are needed to establish EliteTable's relative performance and the contributions of its components.
| Subjects: | Artificial Intelligence (cs.AI); Neural and Evolutionary Computing (cs.NE) |
| Cite as: | arXiv:2610.12536 [cs.AI] |
| (or arXiv:2610.12536v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12536 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Stuart Farmer [view email]
[v1]
Thu, 8 Oct 2026 13:24:21 UTC (762 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →VEPA: Verified Evolving Formal Policy for Lifelong Agent Guardrails
VEPA introduces a framework for verified evolving formal policies for LLM agents, enhancing safety by adapting guardrails based on deployment feedback. It outperforms existing methods across benchmarks like ASSEBench-Safety and ASSEBench-Security, ensuring deterministic policy enforcement even under challenging conditions.