analiza aiqbench
PACT addresses the challenge of training large language models to flexibly switch between multiple diagnostic reasoning paradigms without interference, a critical capability for realistic clinical settings. By combining a privileged data synthesis method with a consensus-based branch training architecture, the framework achieves state-of-the-art performance on a new Chinese medical diagnosis benchmark. This approach offers a scalable solution for building robust medical agents that can handle incomplete patient information while maintaining high diagnostic accuracy.
PACT: Learning Diverse Diagnostic Strategies via Privileged Synthesis and Branch Consensus
arXiv:2606.08938v1 Announce Type: new Abstract: Clinical diagnosis requires flexible use of multiple reasoning paradigms under incomplete patient information. Existing LLM-based medical agents show strong medical reasoning ability, but single-paradigm or naively mixed dialogue supervision makes these paradigms difficult to learn without interference. We propose \textbf{PACT} (Periodic Anchor Consensus Training), a framework that couples supervised multi-paradigm dialogue synthesis with consensus-based Branch training. At the data level, \textbf{DPS} (Doctor-Patient-Supervisor) uses complete electronic medical records (EMRs) for quality control while keeping the doctor agent restricted to patient-visible information. This produces validated dialogues under four diagnostic reasoning paradigms without leaking hidden clinical answers. At the training level, PACT trains one paradigm-specific LoRA Branch per paradigm and periodically aggregates Branches into a shared Anchor through sign consensus. We further construct a dynamic multi-turn Chinese medical diagnosis benchmark for interactive consultation. Experiments show that PACT achieves state-of-the-art performance among compared proprietary, medical-specialized, and task-adapted baselines on diagnostic outcome and consultation-process metrics.
Abstract:Clinical diagnosis requires flexible use of multiple reasoning paradigms under incomplete patient information. Existing LLM-based medical agents show strong medical reasoning ability, but single-paradigm or naively mixed dialogue supervision makes these paradigms difficult to learn without interference. We propose \textbf{PACT} (Periodic Anchor Consensus Training), a framework that couples supervised multi-paradigm dialogue synthesis with consensus-based Branch training. At the data level, \textbf{DPS} (Doctor-Patient-Supervisor) uses complete electronic medical records (EMRs) for quality control while keeping the doctor agent restricted to patient-visible information. This produces validated dialogues under four diagnostic reasoning paradigms without leaking hidden clinical answers. At the training level, PACT trains one paradigm-specific LoRA Branch per paradigm and periodically aggregates Branches into a shared Anchor through sign consensus. We further construct a dynamic multi-turn Chinese medical diagnosis benchmark for interactive consultation. Experiments show that PACT achieves state-of-the-art performance among compared proprietary, medical-specialized, and task-adapted baselines on diagnostic outcome and consultation-process metrics.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2606.08938 [cs.CL]
(or arXiv:2606.08938v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.08938 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 172 cuvinte
Subiecte conexe
Organon : From Evolutionary Search to Recursive Self-Improving Agents for Scientific Discovery
arXiv · 27 oct. 2026
DeepUnfolding_Agent: A Classical_Quantum Adaptive Control API
arXiv · 16 oct. 2026
Unknown Unknowns: Model Misspecification in Machine Learning for Physics
arXiv · 17 aug. 2026
Agentic High-Dimensional Bayesian Optimization with Hypothesis- and Evidence-Guided Search
arXiv · 30 sept. 2026