⚡ LLMsID:
mcts-searchMCTS Reasoning Search
MCTS 推理搜索
🎯Core Definition
MCTS (Monte Carlo Tree Search) organizes generation as tree search: nodes are reasoning states (partial reasoning produced so far), edges are next actions. It loops over four steps — selection (pick nodes by the UCT formula , balancing exploration and exploitation), expansion, simulation (rollout), and backpropagation of the leaf value. With LLMs, node values come from process reward model (PRM) step scores, forming Test-Time Search — the AlphaGo idea transplanted to LLM reasoning.
💡Use Cases
reasoning enhancement on verifiable tasks (math, games), searching high-quality trajectories during reasoning-model training (as RL data), and test-time tree search combined with PRMs.
⚡Key Problems Solved
autoregressive generation goes down one path — a single wrong step is unrecoverable and errors propagate; MCTS explores multiple paths with value-guided pruning, finding better routes at test time and raising solution rates on hard problems.
🎯5 High-Frequency Exam Points
1
What do MCTS's four steps (selection/expansion/simulation/backpropagation) each do?
2
Write the UCT formula, explain each symbol, and how it balances exploration vs exploitation?
3
How is MCTS combined with an LLM? How are nodes, actions, and values defined?
4
The role of PRMs in MCTS? Why are step-level rewards better than outcome rewards?
5
MCTS vs plain sampling (best-of-N): compute complexity vs gains?
📖 In-depth Guide:📄 reasoning-and-cot →
Updated 2026-08-12🎯
🚀 Start Card Practice ➔Test Your Knowledge: Practice Questions for "MCTS Reasoning Search"
Single choice pitfall questions with instant feedback and mistake tracking.