AI Winter Workshop 2024 - CFPU / Brown University Department of Physics (Open to All)

Jan 16, 2024, 3:00 PM → Jan 19, 2024, 5:45 PM America/New_York

Richard Gaitskell (Brown University)

Welcome to the AI Winter Workshop, hosted by the Center for the Fundamental Physics of the Universe, Brown University/Department of Physics

 

All Workshop modules can be viewed here on YouTube.

January 16 - 19, 2024

This Workshop is open to interested graduate students, post-doctoral students and advanced undergraduates from any institution. 

Workshop Goals

The goal of this Workshop is to offer examples of the direct applications of machine learning methods to a range of problems inspired by the use of artificial intelligence in physics. We will provide hands-on experiences with machine learning tools to help reduce the barriers to engagement with these essential tools for physics research. 

Hosted by the Brown Center for the Fundamental Physics of the Universe, this Workshop will consist of six modules consisting of a lecture session delivered by Brown Physics faculty members, followed by practice workshop sessions for participants with direction in the use of machine learning tools.

Register for access to all Workshop materials

A 2-page Summary for each module is available by following the Timetable and selecting the paperclip icon on the module block.

Registration is required for full access to all Workshop materials including notebooks. Please note that we will use your registration email address to grant access to all materials. Full access to all Workshop materials will be available starting January 15. The workshop will make use of Google Colab for all the workbook examples (https://colab.research.google.com/).

If you do not have a Gmail or Google-backed account, you will need to register one.  You can create an account just for this event or perhaps follow the directions at the following link to use an existing email.

Using an Existing Email with Google

Join the AI Workshop Slack for discussion, trouble-shooting, and questions

Support Slack linked here. 

Each module and following workshop are intended to stand alone; participants may sign up for as many modules as they choose. 

All modules will take place virtually. There is no registration fee.

Workshop Material

Below is the link to the workshop material.  Only registrants with a Gmail or Google-compatible email will be able to access it.

Link to workshop Material

 

Citation Guide:

If you chose to use any of the material from this workshop in your own examples, or workbooks, please make a reference back to our workshop, url, and organizers: 
 

AI Winter Workshop - The Center for the Fundamental Physics of the Universe / Department of Physics, Brown University
https://indico.physics.brown.edu/event/2/  
 

Organizing faculty: Richard Gaitskell, Director; Ian Dell'Antonio, Associate Director

A Very Useful Primer in AI ML for All Particpants

Below is a link to a primer that will help you to understand some basic concepts and terminology associated with machine learning.

Machine Learning Primer Link

 

Questions and comments:

Ariel Green (ariel_green@brown.edu)

Tuesday, January 16

Roundtable: What Discoveries and New Techniques will be Enabled by AI?

Join via Zoom here

Moderator: Richard Gaitskell (Brown University)

Respondents:

Sean Carroll (Johns Hopkins University)
https://www.preposterousuniverse.com/
Sean Carroll is a theoretical physicist and is the the Homewood Professor of Natural Philosophy at Johns Hopkins and the author of "Spacetime And Geometry," "The Particle at the End of the Universe," "The Big Picture: On the Origins of Life, Meaning and the Universe Itself," and many other works.

Jesse Thaler (MIT)
https://jthaler.net/
Jesse Thaler is a particle physicist and professor at MIT. He was named the inaugural Director of the NSF Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) in 2020.

David Shih (Rutgers University)
https://sites.rutgers.edu/david-shih/
David Shih is a theoretical physicist and a professor at Rutgers University. His recent research has focused on applications of machine learning and artificial intelligence to data from both colliders and astronomy.

Ilija Nikolov (Brown University)
https://sites.brown.edu/vfm-research-cmpnmr/
Ilija Nikolov is a PhD student and member of the Condensed Matter NMR Group under Professor Vesna Mitrovic. Their research group incorporates AI-designed probes that monitor the quantum phase during magnetic resonance experiments. This gives us an invaluable look into the emergent phases that involve a complex interplay between charge/orbital, spin, and lattice degrees of freedom.

TOPIC:

The growth in the capabilities of AI systems appears exponential, if a little unevenly distributed.

AI is making remarkable new discoveries possible in two ways. On the data discovery side, it has made it possible to sift through the mountains of new experimental data (for example in materials science, biochemistry, particle physics and astrophysics) and find new patterns that would not be noticeable without these techniques. On the theoretical side, it has allowed modelers to narrow down the physical configurations (in quantum systems or in protein folding). What are the next areas that will benefit?

A big unknown remains–when will AI systems be able to come up with their own questions and physics models beyond what humans alone have envisioned? This leap has not yet been made, but it may be coming sooner than we think, or maybe it will turn out that there are types of intellectual inspiration required for scientific progress that machines can never achieve.

Join us for a discussion on how AI is going to change the capabilities of physics and physicists forever.

Conveners: David Shih (Rutgers), Ilija Nikolov (Brown University), Jesse Thaler (MIT), Richard Gaitskell (Brown University), Sean Carroll (Johns Hopkins University)

Wednesday, January 17

Unsupervised learning to find anomalies

Join via Zoom here

We will begin with an executive summary of the current Physics Landscape at the LHC and why the approach and techniques discussed in this module have great potential to discover new physics. Next, we will provide an overview of existing approaches and techniques already been used. The students will develop an unsupervised algorithm using “known” physics processes for its training [2,3]. Then the student will check the validity of this approach utilizing "known anomalies", e.g., jets that are products of two or more merged jets due to a highly Lorentz boosted signatures. The project would provide a modular structure and the possibility of exploring different classifier architectures (e.g., convolutional vs. graph neural networks or regular vs. variational autoencoders) in parallel.

Please register for full access to Workshop materials. Access will be available with the email address you used to register.

Conveners: Greg Landsberg, Loukas Gouskos

2-page Module Summary

Anomaly-Detection-Brown-2024.pdf

Workshop: Unsupervised learning to find anomalies

Join via Zoom here

Conveners: Greg Landsberg, Loukas Gouskos

2-page Module Summary

Link to Module Exercise

Training simulations to predict the first stars and their effect

Join via Zoom here

We are interested in regression tasks in analyzing 21 cm cosmology data. Most semi-numerical algorithms use a unique (and not always physical) parameterization of the 21 cm signal, but we can regress on well-defined phenomenological parameters. In this module, we’ll look at estimating the neutral fraction of the intergalactic medium from 21 cm signal maps.

Please register for full access to Workshop materials. Access will be available with the email address you used to register.

Convener: Jonathan Pober

2-page Module Summary

Workshop Shared Drive

Workshop: Training simulations to predict the first stars and their effect

Join via Zoom here

Conveners: Jonathan Pober, Madhurima Choudhury

2-page Module Summary

Workshop Shared Drive

Thursday, January 18

The evolution of boosted top tagging at the LHC

Join via Zoom here

In this module, we start with an executive summary of jet physics and its importance for the LHC physics program, followed by an overview of the particle tagging techniques developed by the ATLAS and CMS experiments, focusing on the case of boosted, hadronically decaying top quarks. The evolution of ML architectures used for boosted object tagging will be surveyed, beginning from cut-based taggers based on engineered features and proceeding through feature-based networks to the point-cloud, image- and graph-based approaches that are being used in ATLAS and CMS results today.

Please register for full access to Workshop materials. Access will be available with the email address you used to register.

Conveners: Jennifer Roloff, Loukas Gouskos, Matt LeBlanc

2-page Module Summary

Workshop Shared Drive

Workshop: The evolution of boosted top tagging at the LHC

Join via Zoom here

Conveners: Jennifer Roloff, Loukas Gouskos, Matt LeBlanc

2-page Module Summary

Workshop Shared Drive

Galaxy Finding and Sorting in the Vera Rubin Observatory’s LSST and Beyond

Join via Zoom here

In this module we will use the multi-band imaging information and a multi-layer neural network trained on a subset of galaxies with five-band images (u,g,r,i,z, representing images spanning the wavelength to solve a slightly simpler version of the galaxy redshift determination problem. Often, it’s more important to know which galaxies to include or exclude in an analysis as a first step rather than extracting the precise redshift. In this module, you’ll be using a network trained via stochastic gradient descent to separate galaxies into background (galaxies that are gravitationally lensed by the mass) and potential galaxy cluster members.

Please register to receive full access to Workshop materials. Access will be available with the email address you used to register.

Convener: Ian Dell'Antonio (Brown University)

2-page Module Summary

Workshop Shared Drive

Workshop: Galaxy Finding and Sorting in the Vera Rubin Observatory’s LSST and Beyond

Join via Zoom here

Conveners: Ian Dell'Antonio, Shawn Dubey, Tim Launders

2-page Module Summary

Workshop Shared Drive

Friday, January 19

Rare Event Discrimination through Pulse Shape Analysis from LZ Dark Matter Search Experiment

Join via Zoom here

We perform examples of both classification and regression tasks in our analysis of LZ-related data. Classification involves training a machine learning (ML) model [3] to categorize instances according to a particular label for an image (e.g. cat vs dog, signal vs background). Regression involves using an ML model to predict a continuous value (e.g. predicting the length of a dog’s tail). For this module we will be simulated using signals from the LZ detector, rather than pulses from real data events.

Please register to receive full access to module materials. Materials will be accessible with the email address you used to register.

Convener: Richard Gaitskell

2-page Module Summary

Workshop Shared Drive

Workshop: Rare Event Discrimination through Pulse Shape Analysis from LZ Dark Matter Search Experiment

Join via Zoom here

Conveners: Austin Vaitkus (Brown), Richard Gaitskell, Shawn Dubey

Workshop Shared Drive

What electron configurations are possible in materials?

Join via Zoom here

In the presence of strong electronic spin correlations, the hyperfine interaction imparts long-range coupling between nuclear spins. Efficient protocols for the extraction of such complex information about electron correlations via magnetic response are not well known. Here we study how machine learning can extract material parameters and help interpret magnetic response experiments. Although we consider only traditional spin-echo protocols at ideal pulsing, this work develops a framework for testing the efficacy of magnetic probes of electronic features. After applying standard machine-learning techniques to a large dataset of time-series simulations, if the initial parameters are predicted at a rate better than random guessing, then one can surmise that some amount of information is obtained by the proposed experiment. Moreover, a guide for interpreting real measurements is developed from feature ranking of the simulated data.. In this case, we compare the “automatically” generated approach to the previous analytical treatment and highlight improvements that the data-driven approach provides. Our work demonstrates the utility of artificial intelligence in the development of new probes of quantum systems, with applications to experimental studies of strongly correlated materials.

Please register to receive access to full Workshop materials. Materials will be available using the email address you entered to register.

Convener: Vesna Mitrovic

2-page Module Summary

Workshop Shared Drive

Workshop: What electron configurations are possible in materials?

Join via Zoom here

Conveners: Anantha Rao (University of Maryland), Vesna Mitrovic

2-page Module Summary

Workshop Shared Drive