ICTS

I will give a lecture cum hands-on tutorial to demonstrate the core concepts and their implementation. The basic library to be used is PyTorch and PyTorch Geometric. There will be a few coding assignments for the participants throughout the lectures.

Detectors for fundamental physics are among the most complex instruments ever built, and their design today leverages well-established paradigms for the effective reconstruction of particle trajectories and energies from their interactions in suitable media. With the advent of deep learning it is today possible to aim for the full modeling of the physical processes generating the data, the event reconstruction, and the statistical inference, along with the detectors themselves. A differentiable model of the full chain of procedures turning electronic readouts into physics measurements allows for the stochastic gradient descent to the minimum of a loss function that describes the experimental goals, in the space of parameter spanned by construction choices and geometry of the detector. The result is a full optimization of the apparatus, along with understanding of the benefit of each of the construction choices in a continuous multi-dimensional space.

Detectors for fundamental physics are among the most complex instruments ever built, and their design today leverages well-established paradigms for the effective reconstruction of particle trajectories and energies from their interactions in suitable media. With the advent of deep learning it is today possible to aim for the full modeling of the physical processes generating the data, the event reconstruction, and the statistical inference, along with the detectors themselves. A differentiable model of the full chain of procedures turning electronic readouts into physics measurements allows for the stochastic gradient descent to the minimum of a loss function that describes the experimental goals, in the space of parameter spanned by construction choices and geometry of the detector. The result is a full optimization of the apparatus, along with understanding of the benefit of each of the construction choices in a continuous multi-dimensional space.

Detectors for fundamental physics are among the most complex instruments ever built, and their design today leverages well-established paradigms for the effective reconstruction of particle trajectories and energies from their interactions in suitable media. With the advent of deep learning it is today possible to aim for the full modeling of the physical processes generating the data, the event reconstruction, and the statistical inference, along with the detectors themselves. A differentiable model of the full chain of procedures turning electronic readouts into physics measurements allows for the stochastic gradient descent to the minimum of a loss function that describes the experimental goals, in the space of parameter spanned by construction choices and geometry of the detector. The result is a full optimization of the apparatus, along with understanding of the benefit of each of the construction choices in a continuous multi-dimensional space.

TBA

Detectors for fundamental physics are among the most complex instruments ever built, and their design today leverages well-established paradigms for the effective reconstruction of particle trajectories and energies from their interactions in suitable media. With the advent of deep learning it is today possible to aim for the full modeling of the physical processes generating the data, the event reconstruction, and the statistical inference, along with the detectors themselves. A differentiable model of the full chain of procedures turning electronic readouts into physics measurements allows for the stochastic gradient descent to the minimum of a loss function that describes the experimental goals, in the space of parameter spanned by construction choices and geometry of the detector. The result is a full optimization of the apparatus, along with understanding of the benefit of each of the construction choices in a continuous multi-dimensional space.

In this session, we will discuss concepts and practical details of training classifiers for an LHC analysis, with a focus on BDTs. We will also train a BDT classifier implemented using LightGBM.

As even recent developments in machine learning have made their way into the identification of individual objects in high energy physics experiments, and contributed significantly to jumps in physics reach, it is a natural next step to also employ such techniques for the reconstruction of said objects from the raw detector signals. Taking this step poses new and partially unique challenges: conceptually, in terms of resources, and in terms of ensuring robustness and keeping interpretability. This presentation will introduce and motivate these challenges and present first solutions.

Detectors for fundamental physics are among the most complex instruments ever built, and their design today leverages well-established paradigms for the effective reconstruction of particle trajectories and energies from their interactions in suitable media. With the advent of deep learning it is today possible to aim for the full modeling of the physical processes generating the data, the event reconstruction, and the statistical inference, along with the detectors themselves. A differentiable model of the full chain of procedures turning electronic readouts into physics measurements allows for the stochastic gradient descent to the minimum of a loss function that describes the experimental goals, in the space of parameter spanned by construction choices and geometry of the detector. The result is a full optimization of the apparatus, along with understanding of the benefit of each of the construction choices in a continuous multi-dimensional space.

The next generation of ML applications in the physical sciences exploit precise simulations for direct inference. Neural networks are used to estimate high dimensional densities or density ratios, that can then be used for a variety of applications. This section introduces the fundamental concepts of NSBI.

The field of cosmology aims to make transformative discoveries in fundamental physics on multiple fronts ––– the origin of the universe, the nature of dark energy, the nature of dark matter, the mass generation mechanism for neutrinos, modified theories of gravity and beyond. I will present the challenges we are facing in cosmology, and discuss possible venues AI/ML can help.

In this session, we demonstrate how to perform high-dimensional, unbinned unfolding by iteratively estimating likelihood ratios, a procedure known as “Omnifold”. We study it first for a toy problem and then on a physics dataset.

In this session, we will train several deep generative algorithms (GANs, VAEs and NFs) that are typically used in HEP and demonstrate important statistical concepts related to them. The models will be implemented using PyTorch and will be trained using a toy dataset.

Machine Learning is rapidly becoming an important part of nearly all aspects of data analysis in High Energy Physics. This talk will give an overview of Machine Learning in HEP, with a focus on applications in collider physics.

Symmetries are the cornerstones of modern theoretical physics, as they imply fundamental conservation laws. The recent boom in AI algorithms and their successful application to high-dimensional large datasets from all aspects of life motivates us to approach the problem of discovery and identification of symmetries in physics as a machine-learning task. In a series of papers, we have developed and tested a deep-learning algorithm for the discovery and identification of the continuous group of symmetries present in a labeled dataset. We use fully connected neural network architectures to model the symmetry transformations and the corresponding generators. Our proposed loss functions ensure that the applied transformations are symmetries and that the corresponding set of generators is orthonormal and forms a closed algebra. One variant of our method is designed to discover symmetries in a reduced-dimensionality latent space, while another variant is capable of obtaining the generators in the canonical sparse representation. Our procedure is completely agnostic and has been validated with several examples illustrating the discovery of the symmetries behind the orthogonal, unitary, Lorentz, and exceptional Lie groups.

TBA

In this talk, I will describe how the deployment of machine learning to the problem of jet classification unveiled a new phase space of hadronic Higgs boson decays for analysis in CMS, and review the insights gained into the behavior of the Higgs boson due to this advancement. I will then discuss how simulation-based inference techniques may extract even more information about beyond-Standard Model behavior in the Higgs sector.