Inference in Stochastic Optimization with Heavy Tailed Input

Join a SIAM Activity Group on FME Virtual Talk Series presentation where Stanford University's Jose Blanchet delves into stochastic optimization with heavy-tailed input models. Explore compelling empirical evidence from insurance, healthcare, and machine learning sectors demonstrating the effectiveness of infinite variance models in online data-driven decision-making. Examine how these models align with easily monitored historical data features and the operational scales of online algorithms. Learn about newly developed inference tools for monitoring solution quality in infinite-variance stochastic gradient descent (SGD), which expand upon traditional finite-variance analysis and enhance SGD's practical application in online optimization scenarios. Based on collaborative research with Aleks Mijatovic and Wenhao Yang, gain insights into advanced mathematical approaches for handling complex data distributions in real-world optimization problems.