Speaker: Dr Nam Ho-Nguyen, University of Sydney

Title: Risk Guarantees for End-to-End Prediction and Optimization Processes

Synopsis: Prediction methods are often employed to estimate parameters of optimization models. Although the goal in an end-to-end framework is to achieve good performance on the subsequent optimization model, a formal understanding of the ways in which prediction methods can affect optimization performance is notably lacking. We identify conditions on prediction methods that can guarantee good optimization performance. We provide two types of results: asymptotic guarantees under a well-known Fisher consistency criterion and non-asymptotic performance bounds under a more stringent criterion. We use these results to analyse optimization performance for several existing prediction methods and show that in certain settings, methods tailored to the optimization problem can fail to guarantee good performance. Conversely, optimization-agnostic methods can sometimes, surprisingly, have good guarantees. In a computational study on portfolio optimization, fractional knapsack, and multiclass classification problems, we compare the optimization performance of several prediction methods. We demonstrate that lack of Fisher consistency of the prediction method can indeed have a detrimental effect on performance.

WED 07 SEPTEMBER 4PM – 5PM AEST MELBOURNE – IN PERSON

Please attend in-person to support our speaker.

Melbourne Connect 700 Swanston St

Manhari Room, Level 7, 7212

ZOOM MEETING ID: 873 1557 5255; PASSWORD: 778635