Program

The main objective of this course is to learn concepts and implementation of deep learning techniques for scientific and engineering problems. This course entails various methods, including theory and implementation of deep learning techniques to solve a broad range of computational problems frequently encountered in solid mechanics, fluid mechanics, non-destructive evaluation of materials, systems biology, chemistry, and non-linear dynamics. At the end of the course participants will be able to understand and implement

• Neural PDEs: physics-informed neural networks (PINNs): Algorithms, mathematics and implementation in Tensorflow/PyTorch
• Neural operators: DeepOnet:  Algorithms, mathematics and implementation in Tensorflow/PyTorch
• Understanding and using DeepXDE - a library for PINNs and DeepOnet developed at Brown.
• Scalable PINNs and DeepOnet using multi-GPU computing

Modern machine learning has a transformative effect on particle physics research. LHC physics rests on two pillars, huge datasets and precision simulations, which together allow us to answer fundamental physics questions. We will give a brief overview of ML-applications in particle physics and then introduce network-based simulations and simulation-based inference. The key to LHC applications will be control and uncertainty estimates of neural networks.

This tutorial covers a theoretical introduction, as well as a hands-on training session on machine learned potentials and their application in atomistic simulation with quantum mechanical accuracy. We will discuss:

• The basic idea and theoretical foundation of machine learning models
• Molecular representations for machine learning models
• The data generation process based on active learning
• A python-based example, where you can do the whole workflow yourself, covering all steps from data set generation, training and testing of machine learning models, and molecular dynamics simulations

Computer simulations are powerful tools to investigate the physico-chemical mechanisms underlying the structural dynamics of macromolecular systems. However, this task is often exceedingly challenging for conventional simulations, because reorganization events are extremely rare. In these lectures, I will discuss recently developed frameworks that integrate machine learning, molecular dynamics simulations, with quantum computation, to study reorganization events in molecular systems.  Given the exponential growth of quantum computing, approaches like this promise to empower the understanding of how structures and functions emerge from complex molecular systems