Proximal Gradient Methods for Non-Convex Optimization with Applications to Robust Computer Vision

This paper explores the application of proximal gradient methods to non-convex optimization problems, specifically focusing on scenarios arising in robust computer vision. We analyze the convergence properties of these methods under relaxed smoothness conditions and demonstrate their effectiveness in handling outliers and adversarial perturbations. We also discuss practical considerations for implementation and parameter tuning, drawing connections to challenges identified in real-world computer vision deployments. Numerical experiments on synthetic and real-world datasets illustrate the performance of the proposed approach compared to alternative optimization techniques.