Neuro-symbolic (NeSy) AI improves deep learning by integrating reasoning, prior knowledge, and constraints, making it in theory ideal for high-stakes applications. However, this promise depends on learning high-quality abstractions, which is challenging due to potential reasoning shortcuts. This project explores this problem in depth.

Extensible Cognitive Hybrid Intelligence for Deductive Neural Assistance. A neurosymbolic theorem proving platform that transforms Quill (Agda-only neural solver) into a universal multi-prover system with aspect tagging, OpenCyc integration, and DeepProbLog probabilistic logic.

Compare CNN and KAN models on symbolic MNIST addition using DeepProbLog