The lecture is a beginners course in Machine Learning. Nowadays nearly every analysis in physics is based on many variables and their optimal use is a key to get a final result. Each new generation of particle physics experiments is more demanding and finding the signals of new physics become a veritable case of "finding needles in a hay-stack". What helps us in this task are Machine Learning algorithms that give computers the ability to learn without being explicitly programmed.
The aim of the course is to give a basic knowledge about the most popular Machine Learning methods, like Neural Networks and Boosted Decision Trees, and show how they are used in practice. It also gives some mathematical background of the machine learning methods. An important part of the course is an introduction to Deep Learning, a novel machine learning approach developed quite recently and widely used in pattern recognition.
We will use the TMVA package (http://tmva.sf.net) integrated with Root (http://root.cern.ch). So some basic knowledge of C++ and Root would be useful.
Outline:
1. Introduction: introduction to statistics, what does "Machine Learning" mean,? A little bit of mathematics, but also examples of simple linear classifiers.
2. Simple non-linear methods like Naive Bayes, k-Nearest Neighbors, Probability Density Estimators and Boosted Decision Trees (BDT).
3. Neural Networks and Bayesian Neural Networks.
4. Support Vector Machines
5. Cross-validation and optimization of machine learning algorithms. Introduction to Deep Learning.
6. Deep Learning and convolution network. Application of Deep Learning to High Energy Physics problems - Higgs searches.
