Paper 2605.28814
Self-Improving Language Models with Bidirectional Evolutionary Search
- Published
- May 2026
- Research lab
- Independent
- Citations
- 1
- GitHub
- 166 stars
01 In brief
Summary
The paper introduces Bidirectional Evolutionary Search (BES), a search framework for self-improving language models and agents.
BES addresses two limitations of existing methods (best-of-N and tree search): sparse verification signals and confinement to the model's own distribution.
It couples forward search, which uses expansion and four evolution operators (combination, deletion, translocation, crossover) to recombine partial trajectories, with backward search, which recursively decomposes the problem into verifiable sub-goals providing dense feedback.
Theoretical results show expansion-only search is confined to a narrow entropy shell, while evolution operators can escape it, and backward search exponentially reduces the number of samples needed.
Experiments on logical reasoning (Knights-and-Knaves), multi-hop reasoning (MuSiQue), and open problem solving (Circle Packing, Heilbronn) show BES consistently outperforms baselines like GRPO, Tree-GRPO, OpenEvolve, GEPA, and ShinkaEvolve in both post-training and inference settings, with lower variance and modest additional cost.
02 From the paper
Abstract
Search has been proposed as an effective method for self-improving language models and agentic systems, both for post-training sample generation and for inference. However, widely used methods such as best-of-N sampling and tree search face two fundamental limitations: they are guided by sparse verification signals, and they construct candidates primarily through autoregressive expansion, restricting exploration to regions with substantial model probability mass. To address these, we propose Bidirectional Evolutionary Search (BES), a search framework that couples forward candidate evolution with backward goal decomposition. In the forward search, BES augments standard expansion with evolution operators that recombine partial trajectories to generate candidates that are difficult to obtain from a single model rollout. In the backward search, BES recursively decomposes the original task into checkable subgoals, producing dense intermediate feedback that guides forward search. We provide theoretical motivation showing that candidates generated by expansion-only search are confined to a narrow entropy shell while evolutionary operators can escape it, and that backward search can exponentially reduce the number of required samples to find a correct answer. Experiments show that on challenging post-training tasks where mainstream post-training algorithms fail to improve, BES enables consistent gains, and on three open problem solving benchmarks at inference time, BES outperforms existing open-source frameworks in both average and best-case performance. Code and trained models are available at https://github.com/Embodied-Minds-Lab/BES.