Machine-Learning

Objectives

This course introduces the principles of Machine Learning and the most commonly used models.

Course outline

– Principles of machine learning
– Supervised vs. unsupervised learning; training and validation samples, overfitting, generalization error; loss function and error minimization; evaluation of unsupervised methods;
– Neural networks
– Principle of neural networks; properties of simple neural networks; gradient descent; deep neural networks; special architectures (e.g. convolution networks; recurrent networks; …); dimension reduction using neural networks (auto-encoders; word2vec; …).

– Aggregation methods
– Some reminders and more in-depth explanations (CART, multiregression trees), Bagging, random forests, Boosting, XGBoost, Stacking (aggregation of models of different types by construction of a "supervisor" model that best combines the predictions of the primary models).
– Support Vector Machines
– Classification by separating hyper-plane; maximum margin classifier; non-linearly separable data and kernel methods; multi-class SVM; links with other models (logistic, neural networks); gradient descent.

Prerequisites

R, Python, linear algebra, function optimization