NOTE! DIFFERENT LOCATION: OLD CS 2311


Speaker: Yifan Sun

Topic: Optimization methods and graph algorithms for machine learning

Abstract: Optimization is fundamental to training machine learningmodels, yet extending theoretical guarantees to the complex landscapes
of deep neural network training remains a significant challenge. In
this talk, I will highlight key difficulties in fully understanding
optimization in machine learning and present analyses on optimization
flows and higher-order methods, such as quasi-Newton approaches.
Lastly, I will explore how methods from seemingly unrelated domains,
particularly graph-based techniques, can enhance and provide insights
into deep models, underscoring their independent value for further
study.