Home > Events > 2017 Seminars & Colloquia > Dimitris Bertsimas, Massachusetts Institute of Technology

Dimitris Bertsimas, Massachusetts Institute of Technology

Main Content

Machine Learning via a modern optimization lens
30 November 2017 from 4:00 PM to 5:00 PM
201 Thomas Bld
Contact Name
Add event to calendar


The field of  Machine Learning/Statistics (ML/S)  has historically been linked with Probability Theory.

However, some of the central problems of classification and regression     can naturally be written as optimization problems. While continuous optimization approaches has had a significant impact in ML/S   mixed integer optimization (MIO) has played a very limited role, primarily based on the belief that MIO models are computationally intractable.


The period 1991–2015 has witnessed a) algorithmic advances in mixed integer optimization (MIO), which coupled with hardware improvements have resulted in an astonishing 2 trillion  factor speedup in solving MIO problems, b) significant advances in our ability to model and solve very high dimensional robust and convex optimization models.


In this   talk, we demonstrate that modern convex,  robust and especially  mixed integer  optimization    methods, when applied to  a variety of classical ML/S problems can lead to certifiable optimal solutions for large scale instances   that have often significantly improved out of sample accuracy compared to  heuristic methods used in ML/S.


Specifically, we report exciting results on

  1. The classical variable selection problem in regression currently solved by Lasso heuristically.

  2. A systematic approach to design linear and logistic regression models based on MIO.

  3. Optimal trees for classification solved by CART heuristically.


    In all cases we demonstrate that optimal solutions to large scale instances (a) can be found in seconds,   (b) can be certified to be optimal in minutes and (c) outperform classical approaches. Most importantly, this body of work suggests that linking ML/S

    to modern optimization leads to significant advances.


Filed under: ,

Navigation for this Section