Neuro-Symbolic Reasoning
Overview
Neuro-symbolic reasoning studies how to combine neural models’ ability to learn from language, images, and other data with symbolic systems that represent facts, rules, and constraints explicitly. Common approaches use differentiable or probabilistic logic, learned rule induction, and knowledge graphs, or have language models translate problems into programs that solvers, planners, or verifiers can execute. These methods are applied to tasks such as mathematical proof, visual and temporal question answering, robotics, clinical decision support, and policy compliance, where traces and constraints can make decisions easier to check. A central challenge is reliably grounding neural outputs in symbols: translation errors, uncertain evidence, and irrelevant rules can undermine correctness, while stronger verification and structured search may add computational cost or limit flexibility.
Latest papers
279 papers17 papers in the last four weeks, down 35% on the four weeks before. 0.1% of all new papers.