ATPO: Adaptive Tree Policy Optimization for Multi-Turn Medical Dialogue

Researchers developed ATPO (Adaptive Tree Policy Optimization), a new AI algorithm for multi-turn medical dialogues that outperforms existing methods by better handling uncertainty in patient-doctor interactions. The algorithm enabled a smaller Qwen3-8B model to surpass GPT-4o's accuracy by 0.92% on medical dialogue benchmarks through improved value estimation and exploration strategies.

• →ATPO algorithm addresses challenges in multi-turn medical dialogues by formulating them as Hierarchical Markov Decision Processes.
• →The method uses uncertainty-aware adaptive budget allocation to improve value estimation and exploration efficiency.
• →Key optimizations include uncertainty-guided pruning and asynchronous search architecture with KV cache reuse.
• →Qwen3-8B model with ATPO achieved higher accuracy than GPT-4o on three medical dialogue benchmarks.
• →The approach demonstrates significant improvements over conventional RL methods like GRPO and PPO in medical AI applications.