Machine Learning for Physics

13 Mar 2025, 08:50 → 14 Mar 2025, 19:00 Europe/Lisbon

LIP-Lisboa/3-311 - Sala de Seminários (LIP Lisboa)

Cristóvão da Cruz e Silva (LIP), Inês Ochoa (LIP), Michele Gallinaro (LIP), Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

This is a small workshop aimed at interested students to introduce Machine Learning techniques and algorithms applied to HEP. A series of introductory classes using dedicated tools are paired with hands-on exercises to familiarize with the needs and applications adopted for the HEP experiments at the LHC.

The workshop will take place for two full days, Thursday and Friday.

Lectures will be in presence.

 

Please register!

 

This workshop focuses on a specific aspect of the activities at the LHC and is part of the "Course on Physics at the LHC - 2025".

Data challenge!

This workshop includes a data challenge, where the participants will spend an afternoon session trying to solve a given regression problem.

The participants will be divided into groups of three, and the first three groups (according to the metric defined in the data challenge formulation) will receive prizes!!!

1. One copy (per group member) of the book "Deep Learning: Foundations and Concepts", by Christopher M. Bishop and Hugh Bishop. The three copies are courtesy of Springer Nature. The winners will also receive a LIP canvas bag and a LIP paper notebook
2. A LIP canvas bag and a LIP paper notebook (per group member)
3. A LIP canvas bag and a LIP paper notebook (per group member)

First prize sponsored by Springer Nature 

Contact: 

lisbon-ml-workshop@cern.ch

 

The instructors: 

	Pietro Vischia   Ramón y Cajal senior researcher at the Universidad de Oviedo and ICTEA (Spain), Adjunct Professor at IITM. Graduated in 2016 from IST. He is the coordinator of the MODE (Machine-Learning-Optimized Design of Experiments) Collaboration, and the Machine Learning Coordinador of the CMS Experiment at CERN. Specialist in Machine Learning applied to High Energy Physics. Researcher in high-dimensional spaces via gradient descent, eventually powered by quantum algorithms, and on the extension of machine learning methods to realistic neurons with spiking networks, to be then implemented in neuromorphic hardware devices. Within CMS, he focusses on plugging inductive bias in machine learning algorithms for standard model Higgs physics (including the 2018 observation of the ttH process) and beyond-the-standard-model new physics searches in the Top, Higgs, and vector boson sectors. More info at https://vischia.github.io/.
	Inês Ochoa   Researcher at the Laboratory of Instrumentation and Experimental Particle Physics in Portugal. Graduated in 2015 from University College London. Particle physicist in the ATLAS Collaboration, with a focus on searches for new physics via unsupervised learning techniques and developing new algorithms for measuring Higgs boson properties. Expertise in b-tagging and jet substructure, in online and offline systems. Newly appointed co-coordinator of the HEP Software Foundation Reconstruction & Software Triggers group. More info at https://inesochoa.github.io/.
	Cristóvão Beirão da Cruz e Silva   Researcher at the Laboratory of Instrumentation and Experimental Particle Physics in Portugal. Graduated in 2016 from IST. Currently a particle physicist in the CMS collaboration. His research interests focus on detector R&D and the development of precision timing detectors, particularly for the PPS2 upgrade for the HL-LHC. He has additional expertise in data analysis using machine learning techniques, having contributed to the search for the Higgs boson decaying to two photons and SUSY searches with LHC data, particularly the search for the supersymmetric partner of the tau lepton and the search for the supersymmetric partner of the top quark in the compressed mass scenario.

 

 

For the lectures only:

Join Zoom Meeting
https://cern.zoom.us/j/66265392510?pwd=H8rvXWzPBHvQdaYxEOKJMAxP4Qfx2h.1

Meeting ID: 662 6539 2510
Passcode: 073300

 

 

 

 

Thursday 13 March

1 Introduction to the course and welcome

Speaker: Michele Gallinaro (LIP)

2 Introduction to Machine Learning for Physics (lecture)

This lecture will introduce the broad concept of Machine Learning, its connection to Artificial Intelligence, and will broadly review the use of ML in High Energy Physics.

Speaker: Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

2025-03-12_LisbonMLSchoolPhysics_Intro.pdf

HTML slides (preferred, animations work here)

10:00 Coffee Break

3 Automatic Differentiation and Supervised Learning (lecture)

This lecture will focus on supervised learning, a setting where the training data set is “labelled”, that is the target quantity of learning is known. Automatic Differentiation, the technique that powers up modern machine learning frameworks will then be explained in detail, together with its connection to differentiable programming.

Speaker: Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

2025-03-12_LisbonMLSchoolPhysics_SupervisedLearning.pdf

HTML slides (preferred, animations work here)

12:00 Lunch break (lunch not provided)

4 Exercise 1: Network structure and inductive bias in Higgs physics (ttH)

Inductive bias refers to the process of encoding into the learning process some properties of the data known a priori: this can happen by manipulating the training data (augmentation), by modifying the structure of the algorithm (e.g. dense vs convolutional networks), or by modifying the learning target (loss function). The exercise will consist in comparing the performance of generic algorithms with that of algorithms targeted to specific structures (e.g. convolutional networks).

Speaker: Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

Github repository for the course material

15:30 Coffee Break

5 Exercise 2: Classification and anomaly detection in S-top searches

Classification is a category of supervised learning where the goal is to classify the data into different categories. For the CMS search of the supersymmetric partner of the top quark in the compressed mass scenario a Boosted Decision Tree (BDT) algorithm was used to distinguish between signal-like and background-like events. In this exercise, a neural network will be implemented to achieve this task and performance will be compared with the BDT approach. If time allows, a further attempt will be made with an autoencoder neural network, where only background simulated events are used for training, the performance of this approach will also be compared to the previous two approaches. For comparison of the approaches a simplified limit setting via the pyhf library will be used.

Speaker: Cristóvão da Cruz e Silva (LIP)

Github repository for the course material

Friday 14 March

6 Into the belly of Transformers: mathematical formalism and inner workings (lecture)

Transformers are an architecture that powers up most Large Language Models in the market nowadays. This lecture will explain the inner structure of a transformer.

Speaker: Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

2025-03-12_LisbonMLSchoolPhysics_Transformers.pdf

HTML slides (preferred, animations work here)

7 Exercise 3: Flavour tagging with Transformers

Flavour tagging allows us to identify jets that originate from b- and c-quarks, and is a crucial tool for the physics programme of LHC experiments. The jet flavour can be predicted based on the characteristics of the charged particle tracks associated with it. This set of variable number and unordered tracks lends itself to a graph representation, which can be exploited by transformers. In this exercise, we train and evaluate a transformer for identifying b-jets.

Speaker: Inês Ochoa (LIP)

Github repository for the course material

10:30 Coffee Break

8 Unsupervised learning (lecture)

When the data set is unlabelled, that is when the target quantity for learning is not known, traditional supervised learning techniques cannot be used. This lecture will explain the corresponding techniques to obtain learning algorithms without an explicitly known target, such as reinforcement learning.

Speaker: Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

2025-03-12_LisbonMLSchoolPhysics_Unsupervised.pdf

HTML slides (preferred, animations work here)

12:00 Lunch break (lunch not provided)

9 Exercise 4: probing the substructure of boosted jets with unsupervised learning

Speaker: Inês Ochoa (LIP)

Github repository for the course material

15:30 Coffee Break

10 Data challenge!!!

The data challenge will consist in solving a machine learning problem on a given data set.

The participants will be provided access to the data set, and skeleton code to set up the study.

Participants will have to submit a series of predictions for an evaluation data set, as well as the code and an explanation of the logic behind it.

The models faring the best in the evaluation dataset will be declared winner: the first three models will be presented by their developers (1st: 5 minutes, 2nd and 3rd: 2 minutes) during the awards ceremony.

Certificates for the first three classified will be provided.

Speakers: Cristóvão da Cruz e Silva (LIP), Inês Ochoa (LIP), Prof. Pietro Vischia (Universidad de Oviedo and Instituto de Ciencias y Tecnologías Espaciales de Asturias (ICTEA))

17:40 Break (evaluation of challenge outcomes)

11 Wrap-up, Awards, and Group Photo