Neural Reasoning for Scientific and Mathematical Discovery

Neural Reasoning for Scientific and Mathematical Discovery is a forward-looking conference examining how neural network architectures can be designed, trained, and evaluated to perform genuine scientific reasoning—formulating hypotheses, constructing proofs, and discovering mathematical patterns rather than merely pattern-matching on surface features. The conference confronts one of the most important open questions in AI: can neural systems truly reason?

Researchers in machine learning, formal mathematics, automated theorem proving, and AI for science will find the program directly relevant. Sessions address neuro-symbolic integration, reinforcement learning for proof search, mathematical language models, and benchmark design for evaluating reasoning depth. The conference is equally valuable for mathematicians and scientists curious about AI as a research tool and for AI researchers who want their systems tested against the demanding standard of mathematical and scientific rigor.