September 26, 2025
The Institute of Informatics & Telecommunications at NCSR Demokritos, was acknowledged at the Neurosymbolic Learning and Reasoning Conference (NESY 2025), held from 8 to 10 September 2025 in Santa Cruz, California.
Vasilis Manginas, Collaborating Researcher at IIT, and Alessio Lomuscio from Imperial College London, project partners at EVENFLOW, presented their paper A Scalable Approach to Probabilistic Neuro-Symbolic Robustness Verification.
Their work received the conference’s Outstanding Paper Award, bringing IIT’s significant contribution to the forefront of neurosymbolic AI research.
Αbout the paper
The paper introduced a novel method for verifying the robustness of neuro-symbolic (NeSy) systems which combines neural networks with probabilistic logical reasoning.
More specifically, the method can quantify how minor changes in the input of the NeSy system affect the system’s output, thus providing formal guarantees for the safety of the system against adversarial attacks.
Technically, this is achieved by extending robustness verification techniques from the neural to the neurosymbolic setting, facilitated by the fact that NeSy systems can be viewed as end-to-end computational graphs.
