Abstract: We study the worst case complexity of various first order optimization methods for composite minimization and saddle point problems. The nonsmoothness arising in these problems is tackled via*full splitting* methods resulting in easy to implement algorithms. By emphasising the use of stochastic (randomized) methods we are able to present applications in machine learning such as matrix completion and the training of generative adversarial networks.
Join Zoom Meeting: https://us02web.zoom.us/j/83007201160
Meeting ID: 830 0720 1160
Passcode: 2MWNid