Machine learning in physics

Machine learning methods comprise several computational techniques for solving complex problems. Through the use of simple computational units, combined in a neural system, and through a learning system based on the evaluation of selected samples, these systems can solve problems associated with pattern recognition and event classification with better results than traditional methods in many situations of interest.

It is intended that students acquire knowledge and practice in the theoretical grounds of different classification methods in machine learning, such as Neural Networks, Boosted Decision Trees, Random Forest or Nearest Neighbor, useful in different fields of Physics. The student will learn to program and train simple neural network models through interfaces or libraries such as TensorFlow, Apache Mahout, ROOT TMVA, Python mlpy, Keras, or similar, for the production and exploitation of these models.

The level reached by the students will allow them to easily understand the programs written by specialists in the field of pattern recognition in the analysis of detectors, identification and topological classification of events, experimental triggers,... and to design by themselves simple models of machine learning systems.

- Introduction. Basic mathematical concepts (algebra, probability, statistics) and computation for ML.
- Machine learning basis. Neuronal networks: computational units, perceptrons, ...Multilayer architecture. Deep feedforward networks. Convolutional networks. Boosted Decision Trees, Random Forest and Nearest Neighbour methods.
- Learning algorithms. Error function. Supervised and unsupervised learning. Convergence and stability of the solutions. Backpropagation. Training of neural networks using Monte Carlo. Regularization and robustness en deep neural networks. Training optimisation.
- Practicum. Use of ROOT TMVA. Python libraries (mlpy, TensorFlow).

- I. Goodfellow, Y. Bengio, A. Courville, "Deep Learning”, 2016, MIT Press. https://www.deeplearningbook.org/

- A. Geron, “Hands-on Machine Learning with Scikit-Learn and TensorFlow”, 2017, O’Reilly Media. Jupyter notebooks can be downloaded from: https://github.com/ageron/handson-ml

- R. Rojas, "Neural Networks, A Systematic Introduction”, 1996, Springer-Verlag.

- TensorFlow, Open source machine learning framework: https://www.tensorflow.org/

- mlpy, Python module for Machine Learning: http://mlpy.sourceforge.net/

- TMVA:Toolkit for Multivariate Data Analysis with ROOT: https://root.cern.ch/tmva

- J. Torres, "Deep Learning, introducción práctica con Keras", Watch this space 2018, and online https://torres.ai/artificial-intelligence-content/deeplearning/deep-lea…

- C. Albon, "Machine Learning with Python Cookbook. Practical Solutions from Preprocessing to Deep Learning”, 2018, O'Reilly Media.

- Brink, Henrik,"Real-World Machine-Learning", 2016, Manning Publications (Code USC library:3 C10 52)

- Chollet, François, "Deep learning with Python", 2017, Manning Publications (Code USC library: 3 C10 53)

- Carou, David, Sartal, Antonio, Davim, J. Paulo, "Machine Learning and Artificial Intelligence with Industrial Applications", 2022, Springer.

BASIC:
CB6 - Possess knowledge and understanding that provide a basis or opportunity to be original in the development and/or application of ideas, often in a research context
CB7 - Knowledge about how to apply the acquired expertise and skills to solve problems in new or unfamiliar environments within broader (or multidisciplinary) contexts related to the area of study
CB8 - Ability to integrate knowledge and face the complexity of making judgments based on information that, being incomplete or limited, includes understanding on social and ethical responsibilities linked to the application of their skills and judgments
CB9 - Ability to communicate conclusions and the knowledge and ultimate reasons that sustain them to specialized and non-specialized audiences in a clear and unambiguous way
CB10 - Learning skills allowing to continue studying in a way that will be largely self-directed or autonomous.

GENERAL:
CG01 - Acquire the ability to perform work in a research team.
CG02 - Being able to analyze and synthesize.
CG03 - Acquire the ability to write texts, articles or scientific reports according to publication standards.
CG04 - Become familiar with the different modalities used to disseminate results and knowledge in scientific meetings.
CG05 - Apply knowledge to solve complex problems.


TRANSVERSAL:
CT01 - Ability to interpret texts, documentation, reports and academic articles in English, science lingua franca.
CT02 - Develop the capacity to make responsible decisions in complex and/or responsability situations.


SPECIFIC:
CE01 - Know the relevant operating systems and programming languages in physics.
CE03 - Modeling and simulating complex physical phenomena with computers.

There will be expositive (theory) sessions, where the essential concepts and methods are introduced, practical sessions where the students should find the solution to exercises and practical cases in the computer, with the help of the instructors, and tutorial sessions where, either individually or in small groups, particular attention be paid to the students needs and questions. The tutorials will be either presentail or telematic.

Grading will be based on a continuous monitoring to ensure the correct progression in the abilities and knowledge of the students. The instructor will discuss the students solutions to standard practical cases in interactive sessions.

The final grade will be an average of the two with the following weights:
Attending lectures and classroom work: 30%
Completing assignments: 70%

Exceptionally, a final exam may be proposed, if the student has completed all the mandatory assignments proposed during the interactive sessions. In that case the grade will be the maximum between the exam grade and the weighted average of 70% the exam grade and 30% of the continuous evaluation grade.

In case of fraudulent filing of assignments or tests, the provisions of the "Regulations for evaluating the students academic performance and grades review" will apply:

"Article 16. Fraudulent filing of assignments or tests.
The fraudulent filing of any exercise or test required in the grading of an individual will imply the fail qualification in the corresponding call, regardless of the disciplinary process that may be followed against the offending student. It is considered fraudulent, among others, the realization of plagiarized works obtained from sources accessible to the public without re-elaboration or reinterpretation and without citations to the authors or sources. "

Theory: 20 hours (100% attending)
Practice: 10 hours (100% attending)
Tutorial sessions: 1 hours (100% attending presential or telematic)
Personal work and other activities: 44 hours (0% attending)

It is recommended to have a review of the prior basic knowledge of computer programming and Python previously given to the students in other subjects. Personal work and previous experience with machine learning libraries and interfaces will be valuable to better profit from the seminars.