Learning powerful linear models

by Dr Ilya Narsky (Mathworks)

Thursday 1 Mar 2018, 10:00 → 11:30 Europe/Rome

1/1-2 - Aula "C. Voci" (Dipartimento di Fisica e Astronomia - Edificio Marzolo)

In particle physics, supervised learning encompasses techniques such as boosted decision trees and deep neural nets. These techniques learn non-linear models that are accurate yet expensive, hard to visualize and hard to interpret. An alternative approach is to transform data into a higher-dimensional space and learn a linear model in that space.
Although not commonly used in practice, this approach can offer a number of advantages such as simplified parameter tuning, quick training on large datasets and ease of interpretation due to the linear nature of constructed models. In this talk, a linear technique in this class is presented in the context of binary classification.

slides are available at the INFN agenda of this event https://agenda.infn.it/event/18599/