Particle-based methods for non-convex optimization

Speaker: Claudia Totzeck (University of Wuppertal) Title: Particle-based methods for non-convex optimization Abstract: In this talk we discuss the challenges in non-convex optimization and explore the landscape of particle-based methods. In particular, we consider the family of purpose-driven interacting particle methods for global optimization and sampling which originate from Consensus-based optimization. The main focus here is to tailor (stochastic) dynamical systems that are simple enough to allow for rigorous mathematical analysis while on the other hand being rich enough to achieve the task at hand. We will discuss the crucial ingredients of general dynamics such as exploration and exploitation. The theoretical results are underpinned by simulation results and numerical studies. Bio: Claudia Totzeck is currently associate professor at the School of natural sciences and mathematics at University of Wuppertal, Germany. She received her Master degree in Mathematics in Kaiserslautern, Germany. In 2016, she finished her PhD on Asymptotic Analysis and global optimization.  After a postdoc at the University of Mannheim funded by a Margarete von Wrangell Fellowship, she moved to a tenure track position at University of Wuppertal in 2021. Her position was converted to a W2 professorship in 2023. Currently Claudia is vice spokesperson of the CRC 1701 - Port-Hamiltonian systems and her group consists of three PhD students working in the area of optimization with interacting particle systems, network dynamics and optimal control.