Proceedings

Leveraging Over-Parameterization to Improve the Verifiability of Neural Networks

Andrea Gimelli, Luca Oneto, Armando Tacchella

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 580-589.

https://doi.org/10.24963/ijcai.2026/66

Keywords

  • Machine Learning: Deep learning architectures
  • AI Ethics, Trust, Fairnes: Safety and robustness

Abstract

Over-parameterized neural networks, i.e., models with excess

capacity that can fit training data exactly, have demonstrated superior

generalization performance compared to classical models with balanced

capacity. Nevertheless, their deployment in safety-critical domains re-

mains severely constrained by their susceptibility to, e.g., natural per-

turbations and adversarial manipulations. Verification techniques can

solve such problems, but the computational cost of these methods of-

ten scales poorly, specifically when applied to large models. In this work,

we demonstrate that over-parameterization can be exploited not merely

to enhance generalization, but also to mitigate neuron instability, one of

the parameters affecting the efficiency of verification. Our experimental

findings suggest that over-parameterization may serve as a crucial mech-

anism for reconciling the long-standing trade-off between generalization

and verifiability of neural networks.

Andrea Gimelli, Luca Oneto, and Armando Tacchella. 2026. Leveraging Over-Parameterization to Improve the Verifiability of Neural Networks. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 580–589. https://doi.org/10.24963/ijcai.2026/66
Welcome Future Conferences AI Journal Division IJCAI Proceedings About IJCAI IJCAI Awards