{"type":"ScholarlyArticle","title":"Leveraging Over-Parameterization to Improve the Verifiability of Neural Networks","authors":[{"name":"Andrea Gimelli","slug":"andrea-gimelli","given":"Andrea","family":"Gimelli","affiliation":""},{"name":"Luca Oneto","slug":"luca-oneto","given":"Luca","family":"Oneto","affiliation":""},{"name":"Armando Tacchella","slug":"armando-tacchella","given":"Armando","family":"Tacchella","affiliation":""}],"abstract":"Over-parameterized neural networks, i.e., models with excess\r\ncapacity that can fit training data exactly, have demonstrated superior\r\ngeneralization performance compared to classical models with balanced\r\ncapacity. Nevertheless, their deployment in safety-critical domains re-\r\nmains severely constrained by their susceptibility to, e.g., natural per-\r\nturbations and adversarial manipulations. Verification techniques can\r\nsolve such problems, but the computational cost of these methods of-\r\nten scales poorly, specifically when applied to large models. In this work,\r\nwe demonstrate that over-parameterization can be exploited not merely\r\nto enhance generalization, but also to mitigate neuron instability, one of\r\nthe parameters affecting the efficiency of verification. Our experimental\r\nfindings suggest that over-parameterization may serve as a crucial mech-\r\nanism for reconciling the long-standing trade-off between generalization\r\nand verifiability of neural networks.","doi":"10.24963/ijcai.2026/66","url":"https://ijcai.org/proceedings/2026/66","pdf":"https://ijcai.org/proceedings/2026/0066.pdf","markdown":"https://ijcai.org/proceedings/2026/66.md","bibtex":"https://ijcai.org/proceedings/2026/bibtex/66","volume":{"slug":"2026","title":"Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2026","url":"https://ijcai.org/proceedings/2026/","year":2026,"location":"Bremen, Germany","dates":"15-21 August 2026","editors":"Diego Calvanese","isbn":"978-1-956792-09-6","publisher":"International Joint Conferences on Artificial Intelligence Organization","doi_prefix":"10.24963/ijcai.2026"},"section":"AI Ethics, Trust, Fairnes","track":"Main Track","number":66,"pages":{"first":580,"last":589},"keywords":["Machine Learning: Deep learning architectures","AI Ethics, Trust, Fairnes: Safety and robustness"],"videos":[],"references":[],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}