Rutgers University Department of Physics and Astronomy

PHYSICS 693, FALL 2023

MODERN MACHINE LEARNING FOR PHYSICS AND ASTRONOMY


Deep learning and neural networks are actively transforming data analysis in nearly every branch of physics and astronomy. New techniques for data analysis are leading to new discoveries, more sensitive measurements, and faster and more powerful simulations. This course will survey state of the art techniques in modern machine learning and their applications to physics and astronomy.

ML methods covered in this course will include classification/regression, generative modeling and anomaly detection. Students will be introduced to major ML frameworks and architectures such as CNNs, transformers, GANs, VAEs, normalizing flows and diffusion models.

These methods will be illustrated through their applications to a broad array of subfields, such as particle physics, astronomy, cosmology, condensed matter physics, and more. No prior knowledge of modern ML or domain knowledge of any of these fields will be required. However, there will be an emphasis on hands-on examples, so prior experience in data analysis and coding (Python, numpy, matplotlib, ...) would be helpful to get the most out of the course.

Along the way, students will receive a practical introduction to fundamental concepts and methods in machine learning and data science. Hopefully, after taking this semester-long course, students will be able to understand current research trends in ML applications to physics and astronomy and embark on research projects of their own.


ANNOUNCEMENTS

The first meeting of this course will be THURSDAY, SEPTEMBER 7, 2023.

GENERAL INFORMATION

Instructor:

Prof. David Shih
Office: Serin E370
Email: dshih@physics.rutgers.edu
Phone: (848) 445-9072
Office hours: By appointment

Lectures: MTh 10:20-11:40, NHETC SEMINAR ROOM

Resources:

Prerequisites:

This course is intended for everyone in the department, so NO prior domain knowledge of any particular subfield will be required. You will also NOT be required to know in advance any of the deep learning software frameworks (Keras, Tensorflow, pytorch, etc). However, you will be expected to be at least somewhat familiar with python and some of its essential packages (numpy, matplotlib, ...).

Homeworks: TBD

Exams: TBD

Students with disabilities: Please read here.



LECTURE NOTES

  • Lecture 1     Thursday, September 7, 2023
  • Lecture 2     Monday, September 11, 2023
  • Lecture 3     Thursday, September 14, 2023
  • Lecture 4     Monday, September 18, 2023
  • Lecture 5     Thursday, September 21, 2023
  • Lecture 6     Monday, September 25, 2023
  • Lecture 7     Thursday, September 28, 2023
  • Lecture 8     Monday, October 2, 2023
  • Lecture 8     Monday, October 9, 2023
  • Lecture 9     Thursday, October 12, 2023
  • Class presentations     Monday, October 16, 2023
  • Lecture 10     Thursday, October 19, 2023
  • Lecture 11     Monday, October 23, 2023
  • Lecture 12     Thursday, October 26, 2023
  • Lecture 13     Monday, October 30, 2023
  • Lecture 14     Thursday, November 2, 2023
  • Lecture 15     Thursday, November 9, 2023     (SPECIAL GUEST LECTURE: DARIUS FAROUGHY)
  • Lecture 16     Monday, November 13, 2023
  • Lecture 17     Thursday, November 16, 2023     (SPECIAL GUEST LECTURE: DARIUS FAROUGHY)
  • Lecture 18     Monday, November 20, 2023
  • Lecture 19     Tuesday, November 21, 2023
  • Lecture 20     Monday, November 27, 2023
  • Class presentations     Thursday, November 30, 2023
  • Lecture 21     Monday, December 4, 2023
  • Lecture 22     Monday, November 11, 2023

  • DEMOS

  • 10d Gaussian toy model -- DNN binary classifier and Neyman-Pearson comparison (Lecture 5)
  • Top tagging -- cut based and DNN classifier with high level features (Lecture 5)
  • MNIST generation -- vanilla GAN (Lecture 11)
  • GAN Lab demo (illustrates mode collapse) (Lecture 12)
  • MNIST generation -- WGAN-GP (Lecture 13)
  • MNIST generation and latent space visualization -- VAE (Lecture 14)

  • Back to Rutgers Physics Home Page

    Please send any comments on this page to dshih@physics.rutgers.edu.