Teaching Resources

A3D3 members have a range of experience in teaching machine learning courses to a wide education audience. These courses are increasingly drawing on A3D3 research materials for illustrative examples or problem sets. By making these course materials available to the community, we hope to help those educators who are designing or implementing similar courses from the start or looking to update existing courses with fresh material. Queries about individual courses listed below should be addressed to the named course developer.

PHYS 503: Instrumental Physics Applications of Machine Learning

Designed to give students a solid foundation in machine learning applications to physics, positioning itself at the intersection of machine learning and data-intensive science. This course will introduce students to the fundamentals of analysis and interpretation of scientific data, and applications of machine learning to problems common in laboratory science such as classification and regression. There will be two 75-minute classes each week, split into discussions of core principles and hands-on exercises involving coding and data. There will be several projects throughout semester that will build on the course material and utilize open source software and open data in physics and related fields. The list of topics will evolve, according to the interests of the class and instructors. Material will be clustered into units of varying duration, as indicated below. The lists of suggested readings and references are advisory; a large amount of material of excellent quality is now available.

PHYS 460: Machine Learning for Physics

This course presents an introduction to modern data science, artificial intelligence (AI) and machine learning (ML) from a physics perspective. Students learn the basic concepts, tools, and methods of AI/ML applied to scientific challenges. They study how physics knowledge can be incorporated into AI/ML models to improve their learning efficiency, performance, and interpretability. Students also learn how to learn from the machines, including methods for AI explainability and uncertainty quantification. Students explore the different types of learning from data, including supervised, semi-supervised and unsupervised learning. Topics covered include artificial neural networks (NNs), AI/ML-enhanced modeling/simulation, deep generative models, simulation-free inference, variational inference, convolutional NNs, recursive NNs, geometric deep learning, attention mechanism and transformers, auto-encoders, and anomaly detection. Applications to physics will be emphasized, particularly through projects using open scientific data.

PHYS 350: Data Analysis for Physics

This course presents a basic introduction to probability and data analysis from a physics perspective. The methods of extracting meaningful information from data using probability theory and statistical analyses will be presented. Additionally, students will gain familiarity with the concepts through programming exercises using Python notebooks. Topics to be covered include: basics of statistics and probability theory, probability distributions, estimators, uncertainties, confidence intervals and hypothesis testing, Fourier and Monte Carlo methods.

AST 8581 / PHYS 8581 / CSCI 8581 / STAT 8581: Big Data in Astrophysics

This course will introduce key concepts and techniques used to work with large datasets, in the context of the field of astrophysics. In the first 4 weeks of the course, the focus will be on the modern approaches to creating and manipulating large data sets, with the focus on time series analyses and Bayesian methods applied to astrophysics survey data. The remaining part of the course will focus on a range of machine learning techniques for processing data: classification algorithms (supervised and unsupervised learning), clustering algorithms, regression problems, recommender systems, graphic models, and others. The course will dedicate about 2 weeks to each algorithm type: the algorithms will first be introduced in 1-2 lectures, and the emphasis will then be placed on team projects in which the students will apply the algorithms (and already available packages) to astrophysical data sets to answer specific astrophysics questions. The course will assume familiarity with basic concepts in astrophysics, but it will include brief reviews as needed to demonstrate the use of modern data analysis techniques.

PHYS570AI: AI and Physics

The course is designed to be highly interactive, with a mix of foundational and applied machine-learning topics directly applicable to analyzing large scientific data. Students will learn the statistical foundation of machine learning and computing/programming tools for applying ML in research. The first half of the course is dedicated to lectures on the fundamentals of machine learning and popular algorithms, including decision and rule-based methods, deep learning-based models, and others. The lectures will be accompanied by hands-on exercises, fostering a collaborative learning environment. The second half of the course will revolve around a course project, literature discussions, and guest lectures on frontier AI and physics research topics.

Computational Data Science in Physics I

This course aims to present modern computational methods by providing realistic, contemporary examples of how these computational methods apply to physics research. Designed around research modules in which each module provides experience with a specific scientific challenge. In this first module, LIGO data is analyzed, a gravitational wave signal is detected, and this signal is fitted within a physical model, among other objectives, using Jupyter notebooks. Experience in Python helpful but not required. Lectures are viewed outside of class; in-class time is dedicated to problem-solving and discussion. Students taking graduate version complete additional assignments.

Computational Data Science in Physics II

This course aims to present modern computational methods by providing realistic, contemporary examples of how these computational methods apply to physics research. Designed around research modules in which each module provides experience with a specific scientific challenge. Topics included in this second module are hypothesis testing, semi-parameteric methods, and deep learning. In the Final Project, LHC data is analyzed to measure properties of the W boson and Z boson. Experience in Python helpful but not required. Lectures are viewed outside of class; in-class time is dedicated to problem-solving and discussion. Students taking graduate version complete additional assignments.

Machine Learning in Physics

A project-based introduction to machine learning and the application to physics.  Topics include: classification and regression models; neural networks; decision trees. Applications include analysis of high energy physics data and simulation, or other applications.

UCSD PHYS 139/239: Machine Learning in Physics

This course is an upper-division undergraduate course and introductory graduate course on machine learning in physics. No previous machine learning knowledge is necessary. However, some basic knowledge of calculus, linear algebra, statistics, and Python programming may be expected/useful. The course structure will consist of weekly lectures on conceptual topics, e.g. statistics, linear algebra, scientific data set exploration, feature engineering, (stochastic) gradient descent, neural networks, and unsupervised learning. Students will learn key concepts in data science and machine learning, including selecting and preprocessing data, designing machine learning models, evaluating model performance, and relating model inputs and outputs to the underlying physics concepts. We will apply these methods to the domains of collider physics, neutrino physics, astronomy, and potentially others. There will be 4 homework assignments. There will also be a final project in which students will work in groups to reproduce the results of an ML in physics research article. A midterm assignment to propose the project will also be required.

AI/ML Lecture Series at the National Nuclear Physics Summer School

This course is the two-lecture AI/ML series at the NSF-funded National Nuclear Physics Summer School. No prior machine learning knowledge is required, though basic understanding of calculus, linear algebra, statistics, and Python programming may be beneficial. The course consists of two lectures. The first lecture, “AI in a Nutshell: How to Build a Machine Learning Model,” teaches students how to create a machine learning model from scratch to analyze real nuclear physics experiment data. Key concepts such as neural networks, gradient descent, and backpropagation are covered through the process. The second lecture, “Connecting Dots: An AI Cookbook for Nuclear Physics,” builds on the concepts from the first lecture and uses six question/answer pairs to explore a wide range of AI/ML research directions, including geometric deep learning, transfer learning, interpretability, and their applications to solve critical challenges in nuclear physics.