Self-Improving Language Models with Bidirectional Evolutionary Search

• URL: http://arxiv.org/abs/2605.28814v1
• Date: Wed, 27 May 2026 17:59:15 GMT
• Title: Self-Improving Language Models with Bidirectional Evolutionary Search
• Abstract summary: Bidirectional Evolutionary Search (BES) is a search framework that couples forward candidate evolution with backward goal decomposition.<n>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.<n> Experiments show that on challenging post-training tasks where mainstream post-training algorithms fail to improve, BES enables consistent gains.
• Score: 93.21736992722835
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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.

Related papers

• ExTra: Exploratory Trajectory Optimization for Language Model Reinforcement Learning [75.26064451955199]
  ExTra is a framework that extracts exploration signals from the model's own rollouts.<n>It improves Qwen3-1.7B over GRPO by about +5 points on pass@1 and +7 points on pass@16.
  arXiv  Detail & Related papers  (2026-06-23T15:51:39Z)
• Integrating Chain-of-Thought into Generative Retrieval: A Preliminary Study [31.167867050243256]
  ThinkGR is a unified framework that interleaves chain-of-thought with docid generation, enabling iterative thinking and retrieval within a single generative process.<n> Experiments on four multi-hop retrieval benchmarks demonstrate that ThinkGR achieves state-of-the-art performance with an average improvement of +6.86%.
  arXiv  Detail & Related papers  (2026-05-21T11:48:35Z)
• Spend Search Where It Pays: Value-Guided Structured Sampling and Optimization for Generative Recommendation [16.991391135071513]
  We propose V-STAR, a Value-guided Sampling and Tree-structured Advantage Reinforcement framework.<n>V-STAR forms a self-evolving loop via two synergistic components. First, a Value-Guided Efficient Decoding (VED) is developed to identify decisive nodes and selectively deepen high-potential prefixes.<n>Second, we propose Sibling-GRPO, which exploits the induced tree topology to compute sibling-relative advantages and concentrates learning signals on decisive branching decisions.
  arXiv  Detail & Related papers  (2026-02-11T09:57:36Z)
• MMSearch-R1: Incentivizing LMMs to Search [49.889749277236376]
  We present MMSearch-R1, the first end-to-end reinforcement learning framework that enables on-demand, multi-turn search in real-world Internet environments.<n>Our framework integrates both image and text search tools, allowing the model to reason about when and how to invoke them guided by an outcome-based reward with a search penalty.
  arXiv  Detail & Related papers  (2025-06-25T17:59:42Z)
• Constrained Auto-Regressive Decoding Constrains Generative Retrieval [71.71161220261655]
  Generative retrieval seeks to replace traditional search index data structures with a single large-scale neural network.<n>In this paper, we examine the inherent limitations of constrained auto-regressive generation from two essential perspectives: constraints and beam search.
  arXiv  Detail & Related papers  (2025-04-14T06:54:49Z)
• CART: A Generative Cross-Modal Retrieval Framework with Coarse-To-Fine Semantic Modeling [53.97609687516371]
  Cross-modal retrieval aims to search for instances, which are semantically related to the query through the interaction of different modal data.<n>Traditional solutions utilize a single-tower or dual-tower framework to explicitly compute the score between queries and candidates.<n>We propose a generative cross-modal retrieval framework (CART) based on coarse-to-fine semantic modeling.
  arXiv  Detail & Related papers  (2024-06-25T12:47:04Z)
• Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF [82.7679132059169]
  Reinforcement learning from human feedback has emerged as a central tool for language model alignment. We propose a new algorithm for online exploration in RLHF, Exploratory Preference Optimization (XPO) XPO enjoys the strongest known provable guarantees and promising empirical performance.
  arXiv  Detail & Related papers  (2024-05-31T17:39:06Z)
• Learning to Rank in Generative Retrieval [62.91492903161522]
  Generative retrieval aims to generate identifier strings of relevant passages as the retrieval target. We propose a learning-to-rank framework for generative retrieval, dubbed LTRGR. This framework only requires an additional learning-to-rank training phase to enhance current generative retrieval systems.
  arXiv  Detail & Related papers  (2023-06-27T05:48:14Z)
• AugTriever: Unsupervised Dense Retrieval and Domain Adaptation by Scalable Data Augmentation [44.93777271276723]
  We propose two approaches that enable annotation-free and scalable training by creating pseudo querydocument pairs. The query extraction method involves selecting salient spans from the original document to generate pseudo queries. The transferred query generation method utilizes generation models trained for other NLP tasks, such as summarization, to produce pseudo queries.
  arXiv  Detail & Related papers  (2022-12-17T10:43:25Z)
• CorpusBrain: Pre-train a Generative Retrieval Model for Knowledge-Intensive Language Tasks [62.22920673080208]
  Single-step generative model can dramatically simplify the search process and be optimized in end-to-end manner. We name the pre-trained generative retrieval model as CorpusBrain as all information about the corpus is encoded in its parameters without the need of constructing additional index.
  arXiv  Detail & Related papers  (2022-08-16T10:22:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.