1 of 19

Lecture 01

Course Introduction

PHYS 503: Physics Applications of

Machine Learning

Fall 2025

2 of 19

PHYS 503

Fall 2025

3 of 19

Conduct

We strive to promote a productive, inclusive and welcoming environment and require at all times:

  • Treatment of one another with dignity and respect in all forms of interaction and communication
  • The equitable treatment of all members of our community. This includes all axes of human identity related to age, race, ethnicity, nationality, gender identity, sexuality, and physical ability/accessibility.
  • Conduct yourself with the utmost scientific integrity and standards of student and professional ethics
  • Support and encourage each other’s personal and professional growth and well-being
  • Act as needed and appropriate to help maintain an inclusive and welcoming environment for all activities

Furthermore, students may not engage in retaliation against anyone for objecting to a behavior that may violate this code, reporting a violation of this code, or participating in the resolution of such a concern or complaint.

You are required to adhere at all times to the University of Illinois at Urbana-Champaign Student Code found at https://studentcode.illinois.edu

​

PHYS 503

Fall 2025

4 of 19

A Brief History Time and Research…

My research interests are in particle physics and astrophysics

In short, this means I pursue a deeper understanding of the fundamental physics of the universe

PHYS 503

Fall 2025

5 of 19

A Brief History Time and Research…

I started in particle astrophysics studying neutrinos from the Sun within a nickel mine

I moved to Energy Frontier research at the Fermilab Tevatron and more “recently” (15 yrs) at CERN’s Large Hadron Collider

Solution of “Solar 𝝂 Problem”

Neubauer PhD thesis (2001)

PHYS 503

Fall 2025

6 of 19

CERN’s Large Hadron Collider

6

Lake Geneva

Mont Blanc

7 of 19

LHC Experiments

7

ALICE

ATLAS

LHCb

CMS

Lake Geneva

Mont Blanc

The HL-LHC will generate Exabytes of data/year with increased complexity

8 of 19

How do we look for new physics at colliders?

8

9 of 19

How do we look for new physics at colliders?

9

10 of 19

10

  • Higher luminosity running leads to increase size and complexity collision data

→ a serious challenge for detector triggering and event reconstruction in the experiments during HL-LHC (<μ>~200) running!

Collision event recorded by the CMS detector during a high luminosity running of the LHC with <μ> ~ 100

11 of 19

Typical LHC Data Flow → HL-LHC

11

Collisions

Level 1 Trigger

99.75% of events rejected!

High Level Trigger

99% of events rejected!

​

Offline Reconstruction @ Tier-0

40 MHz

O(100 TB/s)

1 kHz → 1 MHz

1 kHz → 10 kHz

1 MB → 5 MB / event

Lots of human & artificial intelligence steps here!

99.9975% of collision events are rejected while retaining those essential for our science!

  • Must continually choose a winner out of 40,000 (on avg.) collisions very wisely and must choose it very fast!

Discovery using Higgs!

12 of 19

Data Deluge: Triggering at LHC

12

13 of 19

Deep Learning Revolution

13

  • Many scientific challenges can be addressed by mapping them to tasks common in artificial intelligence and machine learning
  • Rise of physics-inspired neural networks in particle physics (e.g. Lorentz-invariant NNs. …)

Work in progress…

14 of 19

Deep Learning Revolution

14

  • Many scientific challenges can be addressed by mapping them to tasks common in artificial intelligence and machine learning
    • E.g. Identification and measurement of jets from energy deposits in calorimeters at particle colliders can be cast as an image classification problem

1603.09349

W🡪qq’

QCD jet

Average over 100,000 jets

Typical jet in an event

15 of 19

Deep Learning Revolution

15

  • Many scientific challenges can be addressed by mapping them to tasks common in artificial intelligence and machine learning
    • E.g. Identification and measurement of charged particle trajectories in a tracking detector can be cast as a geometric deep learning problem

Node

Edge

 

 

 

 

 

 

 

 

 

INTERACTION NETWORK

Graph Neural Network

16 of 19

Data Deluge: Triggering at LHC

16

https://a3d3.ai

https://fastmachinelearning.org

17 of 19

AI meets FPGAs

17

Jigglypuff

18 of 19

18

https://sites.google.com/view/hls4mlsummerschool/home

19 of 19

Course Introduction

https://github.com/illinois-ipaml/MachineLearningForPhysics

​

PHYS 503

Fall 2025