Deadline for registration: 15 days before the starting date of each training
Duration : 4 days / (28 hours)
This training course enables the participants to reinforce their theoretical and practical knowledge in order to implement machine learning techniques for the automatic analysis of data. The main statistical methods for data analysis are presented, both for data exploration (non-supervised learning) and for prediction (supervised learning). Each method is first presented and commented on a theoretical level, and then illustrated on numerical experiments run with public datasets using R and/or python/scikit-learn software.
Objective of the training
To know the main algorithms of automatic data analysis, and to know how to use them with R and/or python/scikit-learn.
The participants should be able to :
- recognize the type of problem that they are facing (supervised or non-supervised learning, sequential learning, reinforcement learning…);
- choose the right algorithm to use;
- use an R on python implementation of this algorithm.
This training session is for students, engineers, and computer scientists who wish to reinforce or extend their theoretical background and practical knowledge on automatic data analysis by statistical learning algorithms.
Basic knowledge in statistics: elementary probability, statistical tests, Gaussian linear model.
Basic knowledge in algorithmic and programming.
Install Python 2.7 with Anaconda, R 3.4.2 and IRkernel. Internet access during the sessions in order to get possible updates or to load additional libraries.
In order to verify that the prerequisites are satisfied, the following questionnaire must be completed. You need to get at least 75% of correct answers in order to be authorized to follow this training session. If you don’t succeed it, your subscription will not be validated. You only have two chances to complete it.
Questionnaire 1 https://goo.gl/forms/xL86TzPDFOC5r7ln1
Scientific contacts: Philippe Besse, Sébastien Gerchinovitz, Béatrice Laurent-Bonneau
- Traines/PhDs/PostDocs : 280 €
- Cerfacs shareholders/CNRS/INRIA : 800 €
- Public : 1600 €
Every day from 9h to 17h30.
Morning: lecture; afternoon: hands-on sessions.
General presentation of statistical machine learning and its main approachs. Comparison with traditional statistics and machine learning.
– Principal component analysis
– Agglomerative Hierarchical Clustering
– k-means, k-medoids and variants
– overview of other methods : Affinity Propagation, dbscan, etc.
Supervised learning 1 / 2 :
– k nearest neighbors
– Gaussian linear model, logistic regression, model selection
– LASSO et variants
– Support Vector Machines
Supervised learning 2 / 2 :
– Decision Trees
– Bagging, Random Forests, Boosting
– Neural networks, deep learning
Sequential learning, multi-armed bandit problems
Super-learning and expert aggregation
Reinforcement learning (introduction)
A final exam will be conducted during the training.