New Research Explores Reinforcement Learning Methods for Clinical Decision-Making and AI Agent Training

New Research Explores Reinforcement Learning Methods for Clinical Decision-Making and AI Agent Training

Researchers have published multiple papers advancing reinforcement learning (RL) techniques for complex AI applications, including clinical decision-making and agent training.

According to arxiv.org, a new study formulates oncology treatment planning as an interactive RL problem under partial observability. The researchers used GRPO and GDPO to optimize Llama-3.1-8B-Instruct, constructing “patient-specific, process-aware rewards from decision processes observed in real world clinical records.” The GDPO-trained policy increased Mean@3 scalar clinical reward by 79.60% over base Llama and exceeded GPT-5.2 by 18.82% under the same interactive environment.

In separate work on tool orchestration, arxiv.org reports that researchers introduced ProgAgent, which uses Programmatic Tool Calling (PTC) where “the model invokes tools through executable code and stores intermediate results in runtime variables rather than appending them to the context.” Using only 215 tasks, ProgAgent-14B achieved a 30.7% pass rate at a 64k-token environment state on LOCA-bench, where comparable baselines scored zero.

Additionally, arxiv.org published EventLens, a “vision-centric reinforcement learning framework that learns video understanding through event-structure recovery,” and a comprehensive survey of Flow Matching in RL, which arxiv.org describes as “a principled and efficient generative modeling framework for reinforcement learning” that uses “deterministic probability transport” and “sampling based on ordinary differential equations (ODEs).”