Semialgebraic representation of monotone deep equilibrium models and applications to certification

Published in Advances in Neural Information Processing Systems (NeurIPS), 2021

This paper proposes three general models for verifying robustness of monotone deep equilibrium models (monDEQ). Related approaches include the pioneer one by [Pabbaraju et al., 2021] using Lipschitz constant, and the SOTA by [Müller et al., 2023]. [bib, link, code]