An unofficial implementation of the method introduced in the paper [1]. Also used as a baseline in the papers [2, 3]. Requires NumPy and CvxPy. The MOSEK solver is recommended.
[1]. Neural Networks are Convex Regularizers: Exact Polynomial-time Convex Optimization Formulations for Two-layer Networks. Mert Pilanci, Tolga Ergen. Proceedings of the 37th International Conference on Machine Learning, 2020.
[2]. Practical Convex Formulations of One-hidden-layer Neural Network Adversarial Training. Yatong Bai, Tanmay Gautam, Yu Gai, Somayeh Sojoudi. American Control Conference, 2020.
[3]. Efficient Global Optimization of Two-layer ReLU Networks: Quadratic-time Algorithms and Adversarial Training. Yatong Bai, Tanmay Gautam, Somayeh Sojoudi. SIAM Journal on Mathematics of Data Science, 2022.