Structured Reasoning
Overview
Structured reasoning aims to make AI systems solve multi-step problems by organizing information into explicit plans, logical rules, graphs, tables, or evolving states. Common approaches include chain-of-thought and hierarchical planning, symbolic constraints and program execution, structured memory, and step-level verification or recurrent refinement; training uses curated traces, supervised fine-tuning, and reinforcement learning. Systems are tested on logic puzzles, mathematics, multi-hop questions, structured tables, and visual or scientific tasks, where performance often declines as problems grow deeper or inputs become harder to parse. A central challenge is ensuring that a correct final answer follows from valid, evidence-grounded steps, while keeping the reasoning reliable and efficient across tasks and model sizes.
Latest papers
86 papers9 papers in the last four weeks, up 125% on the four weeks before. 0.1% of all new papers.