Séminaire en format hybride au GERAD local 4488 ou Zoom.

In this talk, we will address the online dynamic stochastic optimization problems with quasar-convex loss functions, a class that has recently been demonstrated to be satisfied for applications such as the identification of linear dynamical systems and generalized linear models. We utilize online gradient descent and derive regret bounds based on cumulative path variation and cumulative gradient variance. These results are then applied to problems including generalized linear models, phase retrieval, and tomographic reconstruction. Lastly, numerical experiments are presented to corroborate our theoretical findings.

Bio: Yuen-Man Pun is currently a postdoctoral fellow with the CIICADA Lab, School of Engineering at Australian National University. Prior to that, she received a PhD degree in Systems Engineering and Engineering Management (SEEM) in 2022, an MPhil degree in SEEM in 2018 and a BSc degree in Mathematics in 2016, all from the Chinese University of Hong Kong. Her research focuses on optimization methods, algorithmic design and analysis, and its applications in data science and signal processing.