Li Xiangtian

Li Xiangtian

he/him

Researcher focusing on Large Language Model Post-training, Reinforcement Learning, Multi-agent Systems, and Multimodal AI. Previously worked on Embodied AI.

MLLM-MCTS-CoT: Monte Carlo Tree Search for Multimodal Reasoning

发布时间：

Overview

This project integrates Monte Carlo Tree Search (MCTS) algorithms with Chain-of-Thought reasoning to improve the reasoning capabilities of multimodal large language models, enabling better exploration of reasoning paths.

Key Features

  • MCTS Integration: Applying Monte Carlo Tree Search to reasoning tasks
  • Enhanced Reasoning: Better exploration and exploitation of reasoning paths
  • Multimodal Support: Works with both visual and textual inputs
  • Search Strategy: Intelligent search through reasoning space

Technologies

  • PyTorch
  • Monte Carlo Tree Search (MCTS)
  • Multimodal Large Language Models
  • Reinforcement Learning Concepts