ML under a Modern Optimization Lens

Monday, September 16, 2019 - 9:00am - 10:00am
Lind 305
Dimitris Bertsimas (Massachusetts Institute of Technology)
We present three examples from central problems in machine learning: sparse regression, stable regression and matrix completion. We utilize discrete and robust optimization to demonstrate that using modern optimization we can find solutions to large scale instances of these problems that
(a) can be found in seconds/minutes.
(b) can be certified to be optimal in minutes/hours.
(c) outperform classical heuristic approaches
in out of sample experiments involving real world and synthetic datasets.