Large-scale Machine-Learning

Objectives

– The transition to large-scale databases is changing certain uses of statistical learning. We’ll look at a few examples, with the underlying theoretical justifications.

– Objective 1: be able to choose appropriate approaches for a given "large-scale" problem, which requires ..

– Objective 2: to know how the approaches presented behave in terms of storage size and/or computation time and/or estimation quality.

Course outline

– Stochastic gradient descent

– Data and parsimonious models (low-rank matrices, random Fourier attributes, distillation, …)

– Distributed computing

Prerequisites

Big data panorama, machine learning, statistical learning, optimization