DElenchus
Neuro-Symbolic AI for Safety-Critical Cyber-Physical Systems
NeurIPS 2026 — in preparation
A physics-grounded agent that treats safety as a structural invariant rather than a reward signal. Built on Monotonic Neural ODEs and Active Inference with Hierarchical Markov Blankets.
Overview
DElenchus is a physics-grounded agent that treats safety as a structural invariant rather than a reward signal. It is built on Monotonic Neural ODEs and Active Inference with Hierarchical Markov Blankets, and targets fault localization and epistemic refusal in systems governed by physical law.
A submission to NeurIPS 2026 is in preparation.