R-APS: Compositional Reasoning and Meta-Learning for Constrained Design

Arxiv CS.AI (top papers)

R-APS uses reflective adversarial Pareto search to enable LLMs to handle constrained design tasks through compositional reasoning and in-context meta-learning. The approach addresses the gap between LLM fluency and reliable agentic performance in extended-horizon tasks.

llm

Arxiv CS.AI (top papers)·Jo\~ao Pedro GandarelaThiago RiosStefan MenzelAndr\'e Freitas

7 days ago