Practical Optimization for Real-World Statistical Problems

• Activity Number: 98
• Type: Roundtables
• Date/Time: Monday, August 10, 2015 : 7:00 AM to 8:15 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Glen Colopy*
• Companies: University of Oxford
• Keywords: Optimization ; Non-convex ; Machine learning ; Bayesian ; Sampling ; Prediction

Abstract

Optimization pervades many statistical techniques, from selection of the best predictive model from a set of candidates, to maximum-likelihood parameter search, to posterior sampling techniques, and in selecting between competing performance metrics. Optimization also underlies many non-probabilistic machine learning techniques, such as kernel machines. While non-convex optimization algorithms are covered extensively in Operation Research programs, few statisticians have formal training in optimization algorithms and instead pick up techniques as they go. This round table will allow attendees to share knowledge of software and learning resources that have assisted them in applied optimization problems, and describe current hurdles they're facing in applying optimization to their statistical work. The round table will begin with a quick survey of interesting recent applications, with attendees encouraged to contribute experiences from their own work. Finally, suggested learning material and technical resources will be covered for participants looking to get started in optimization or to improve upon their current proficiency.