Tag
This paper introduces ExTS, a tree-search policy for budget-constrained agentic search in LLM agents that improves over standard baselines in tasks like prompt optimization and code generation with an average +5.5% gain.
This paper identifies common failure modes in tree-search-based autoresearch agents applied to tabular datasets, such as repeated bug resolution, poor hyperparameter tuning, and ineffective exploration, and proposes targeted interventions like a global debug consultant and refined tree-search algorithms to recover wasted compute and improve performance without changing the underlying language model.
A tweet from DAIR AI highlights new research by Sakana AI introducing CEDAR, an LLM-agent method that designs, simulates, and refines complex system-dynamics models via Monte Carlo Tree Search over feedback structures, aiming to automate goal-directed design in artificial life.
SkillHEX proposes a closed-loop framework for autonomous skill evolution in LLM agents, using hypothesis-driven self-verification and evidence-guided tree search to overcome sparse reward challenges. It outperforms existing self-evolving methods on SkillsBench with limited interaction budgets.
GATS introduces a Graph-Augmented Tree Search with a layered world model (symbolic, learned, generative) to eliminate LLM calls during planning, achieving 100% success on synthetic tasks and stress tests, outperforming LATS and ReAct.
Introduces SVA, a framework that decouples action generation from consequence evaluation in frozen VLA models using Monte-Carlo tree search and distillation into a lightweight Q-value model, improving generalization and task success rates while reducing computational costs.
SproutRAG is a hierarchical RAG framework that uses attention-guided tree search and progressive embeddings to retrieve at multiple granularities from long documents, improving information efficiency by 6.1% over baselines.
LLMZero uses LLM agents to search over training trajectories via tree search, discovering adaptive multi-parameter transitions for RL post-training that outperform fixed schedules and grid search across diverse tasks.
A new Nature paper introduces ERA, an AI system that iteratively writes, runs, scores, and improves scientific code through tree search, moving AI for science from text generation to code testing.
This paper introduces Collective Skill Tree Search (CSTS), a framework that constructs structured, diverse, and generalizable trees of skills for LLM agents using collective intelligence from multiple models. The resulting model, OpenClaw-Skill, demonstrates improved agentic capabilities in long-horizon planning, tool use, and generalization.
StarOR proposes a framework that synergizes Monte Carlo Tree Search with test-time reinforcement learning for automated optimization modeling, achieving state-of-the-art performance across multiple benchmarks.
Arbor introduces structured tree search as a cognition layer for autonomous agents, enabling multi-day, full-stack LLM inference optimization with up to 193% throughput-latency improvement over vendor baselines through a checks-and-balances multi-agent architecture.
MLEvolve is a self-evolving LLM-based multi-agent framework for automated ML algorithm discovery that extends tree search to Progressive MCGS with graph-based cross-branch information flow and retrospective memory. It achieves state-of-the-art performance on MLE-Bench and outperforms AlphaEvolve on mathematical algorithm optimization tasks.
ATOM is a multi-agent framework that formulates molecular optimization as a tree-structured search with specialized agents along paths, enabling exploration of alternative molecular trajectories and improving Pareto coverage in multi-objective benchmarks.
This paper theoretically studies how transformer-based policies acquire search capabilities from reinforcement learning training dynamics in a stochastic tree environment. It shows that a two-head transformer can implement depth-first search and that this mechanism emerges naturally from sparse reward signals under a depth-wise curriculum.
This paper proposes three rerooter designs for Levin Tree Search that leverage state-space structure and learned heuristics to improve search efficiency without explicit subgoal generation, achieving state-of-the-art online training efficiency.
The paper proposes 2FFS, a two-fidelity tree-search algorithm that adaptively balances cheap biased evaluations with expensive accurate evaluations in stochastic minimax trees for fixed-confidence best-action identification, with theoretical guarantees and experimental efficiency gains.
Guowei Xu discusses limitations of Best-of-N and tree search methods for LLMs on hard reasoning problems, noting sparse verification signals and that candidates remain within the model's distribution.
This paper introduces MAPLE, a tree search method that aggregates policy and value evaluations from multiple sampled world states, extending AlphaZero to imperfect-information games. Experiments on Phantom Go and Dark Hex show Elo improvements of 291 and 136 over the PIMC-based AlphaZero baseline.
This paper introduces AutoMMemo, a framework that enables multimodal agents to automatically design memory mechanisms (expressible as executable memo programs) for learning from multimodal interaction trajectories, outperforming no-memory and fixed-memory baselines on GUI/Web navigation and visual reasoning benchmarks.