AI4EIC-2025 (4^th workshop)

C. Fanelli, T. Horn on behalf of AI4EIC

https://eic.ai/

4^th AI4EIC Workshop

Hosted by MIT/IAIFI

2

Thanks to Organizing Committees (AI4EIC-2025)

3

Local Organizing Committee (MIT):

• Ethan Cline
• Gian Michele Innocenti
• Marisa LaFleur
• Mike Williams
• Noah Wuerfel
• Or Hen
• Phiala Shanahan
• Thomas Bradford

Special thanks to Marisa!!

Local point of contact: mlafleur@mit.edu

MIT

NSF AI Institute for Artificial Intelligence and Fundamental Interactions (IAIFI)

Thanks to Organizing Committees (AI4EIC-2025)

4

Scientific Organizing Committee:

• Abhay Deshpande, BNL/SBU
• Ben Nachman, Stanford
• Cris Fanelli, W&M
• David Lawrence, JLAB
• Malachi Schram, JLAB
• Marco Battaglieri, INFN

• Mike Williams, MIT
• Or Hen, MIT
• Phiala Shanahan, MIT
• Tanja Horn, CUA
• Torre Wenaus, BNL

AI4EIC Contact:

​

ai4eic@gmail.com

support@eic.ai

Email support@eic.ai to subscribe to mailing list

Check-out ai4eic Slack channel on eic.ai

Resources

5

• Wifi Access
• Eduroam or MIT Guest (provide an email or phone number, then verify)
• AI4EIC-2025 Workshop Indico
• http://indico.bnl.gov/event/AI4EIC2025
• AI4EIC Website
• https://eic.ai

​

• We are aiming to share recordings of as many of the talks as possible. Please use this link or this QR code to fill out the consent form: http://bit.ly/4hz6u5w

​

Meeting Location & Other Information

6

• MIT, Wong Auditorium (Tang Center)
• Tang Center, 70 Memorial Drive, Cambridge, MA 02138
• All sessions will also be available on Zoom: https://mit.zoom.us/j/99550594447
• Parking near MIT is available using public parking garages: https://web.mit.edu/Facilities/transportation/parking/visitors/public_parking.html
• The closest one to the Tang Center is the One Memorial Drive Garage.

• If you are traveling to MIT via public transportation, you can take the red line to Kendall/MIT (and then it is about a 5-minute walk).

Meals/Coffee-Breaks

7

• There will be coffee/tea available in the room every day
• On Day 1—Monday, October 27— we will host a reception with food and beverages from 6:00 to 7:30 pm in the Samberg Conference Center (Salon M on the 7th floor), which is right next door to the Tang Center. We hope you can join us!
• Lunch/Dinner: on your own —
• We’ve compiled a list of suggestions for places to eat (at lunchtime or otherwise) and things to do around Cambridge/Boston if you are getting into town early or have time in the evening: https://docs.google.com/document/d/1iw0lDBL6kVCW5vhaaxZtoYfxnWEbZpJE/edit?usp=sharing&ouid=107650713859084891843&rtpof=true&sd=true.

Format: 2 sessions/day for 3 Days (Oct 27-29)

8

Day 1

Day 2

Day 3

Day 1 (Oct 27)

9

AI/ML for Accelerators

Calib, Monit, Exp. Control

Lunch break: 13:15-14:30

Reception: 18:00-19:30

Day 2 (Oct 28)

10

AI/ML for ePIC & Beyond

Lunch break: 13:15-14:30

AI/ML for Data Analysis & Theory

9:00-10:00 RAG4EIC Tutorial (K. Suresh/AI4EIC)

Day 3 (Oct 29)

11

Trends in Data Science

Lunch break: 13:15-14:30

Distributed ML/ Production

9:00-10:00 RAG4EIC Tutorial (K. Suresh/AI4EIC)

Tutorial: RAG4EIC Agent

12

12

What is Retrieval Augmented Generation (RAG)?

• Access up to date information without explicitly training of LLM.
• Reduce “Hallucination” of LLM.
• Grounding LLM to truth to increase reliability by providing citations.

Why need RAG for Large Scale Physics Experiments?

• EIC large scale experiment (e.g., EICUG ~1,500 users, ePIC 180 institutions)
• Regular updates to documents, Wiki etc; Tot document size ~ scale of experiment
• Newbies may take months to get to know the full experimental details

“Ingestion” of data

• Creation of vectorized knowledge base
• Every node below influence RAG performance
• 200+ recent arXiv papers on EIC (since 2021)

“Inference”

• Given a prompt compute similarity index to most similar vectors in VectorDB
• Use LLM to further narrow down and summarize the finding

(source code) https://github.com/ai4eic/EIC-RAG-Project

(agent) https://rags4eic-ai4eic.streamlit.app/RAG-ChatBot

K. Suresh, et al "Towards a RAG-based summarization for the Electron Ion Collider." JINST 19.07 (2024): C07006.

ATLAS (LHC) also working in this direction. See ChATLAS, D. Murnane et al, presented at AI4EIC,

https://eic.ai/ai4eic-resource-hub

http://rag4eic.ds.wm.edu

Tutorial: RAG4EIC Agent

​

13

• AI4EIC Collaboration with Ramaiah University, Data Science (India)
• B.Sc. thesis by Tina J Jat (supervised by Tapasi Ghosh, in collaboration with AI4EIC) – This project advances prior OpenAI-based work by developing a fully open-source RAG pipeline for the EIC, using LLaMA 3.2/3.3 and mxbai-embed-large. With a local vector database, it ensures data privacy, avoids external sharing, and remains cost-effective.
• Now EIC RAG works with open source models within olamma. We are working on moving towards agentic workflow
• Topical Meetings: a topical meeting will take place between July 30, 2025, 11am EST (https://indico.bnl.gov/event/28948/)
• Tutorials@AI4EIC2025: Tutorials on AI agents for EIC will take place during the workshop on Day 2 and Day 3.

(source code) https://github.com/ai4eic/EIC-RAG-Project

(agent) https://rags4eic-ai4eic.streamlit.app/RAG-ChatBot

Community Effort

https://eic.ai/ai4eic-resource-hub

AI4EIC Workshops

14

https://eic.ai/workshops

• 5 workshops organized or co-organized by the AI4EIC WG. In the last 12 months we contributed to organize the joint session with the Streaming Readout at the Streaming Readout Workshop SRO-XII held in Tokyo Dec 2024, and we are currently organizing the 4th AI4EIC workshop that will be held at MIT supported by the Institute for Artificial Intelligence and Fundamental Interactions (IAIFI)

Past Editions

MIT/IAIFI (2025)

CFNS (2021)

W&M (2022)

CUA (2023)

+AI4EIC (2024)

40 contributions overall, with 1:4 on AI/ML in SRO.

All contributions to AI4EIC-2025 can be published as proceedings on JINST

AI4EIC Publications

15

https://eic.ai/living-review

14 citations (INSPIRE-HEP)

2022*

2024*

2024

*(not full list; 20+)

More information will be provided after the workshop on how to submit the AI4EIC-2025 proceedings on JINST

DOE Recognition of AI4EIC

16

• The AI Working Group within the EICUG has gained notable recognition in recent DOE Funding Opportunity Announcements, underscoring the growing impact of our efforts on supporting the future of EIC science with AI/ML
• Highlighted in two consecutive FOAs DE-FOA-0002875 (2023) and DE-FOA-0003845 (2025) both reference the AI4EIC workshops, emphasizing our role in addressing how AI/ML might contribute to advance the research, design, and operation of the future EIC.
• Recognition reflects the DOE’s alignment with EICUG’s vision to integrate AI/ML with experimental data analysis and high-performance computing in nuclear physics.

DE-FOA-0002875 (deadline: Jan 11, 2023; award ceiling $2M/year)

Artificial Intelligence and Machine Learning For Autonomous Optimization and Control of Accelerators and Detectors

​

DE-FOA-0003845 (deadline: Jan 14, 2025; lab award ceiling $3.5M/year)

Artificial Intelligence and Machine Learning Applied to Nuclear Science and Technology

These mentions affirm our trajectory and provide momentum for continued cross-disciplinary collaboration and future funding opportunities.

Spotlight on AI/ML for EIC Science

17

Full list can be found here: https://eic.ai/spotlight

• This spotlight features a curated selection of the most influential papers per year that showcase cutting-edge applications of AI and ML to EIC science.
• These works represent milestones in the integration of modern computational techniques into experimental and theoretical efforts within the EIC community. We highlight these papers for their innovation, impact, and potential to guide future research in EIC.

Impact of AI4EIC Initiatives: Examples

18

AI4EIC hackathon

(ML for Cherenkov detectors)

(2022)

Inspired development of ML algorithms for Cherenkov detectors

dataset

CHEP

proceeding

(2024)

Mach. Learn.: Sci. Technol. 6 015028

(2025)

AI4EIC hackathon

(LLM for EIC Science)

(2023)

(2024)

JINST

proceeding

RAG4EIC Project

JINST 19.07 (2024): C07006.

https://github.com/ai4eic/EIC-RAG-Project

The seeds planted through AI4EIC initiatives continue to bear fruit over the years, giving rise to follow-up research and publications

Live Notes

19

• Link to Live Notes is here
• Everyone can edit
• Conveners can use it to make notes
• Live Notes are useful for attendees to write down their questions;
• Speakers can reply to any pending questions after their talks

​

Thank You & Welcome to AI4EIC 2025

20

• Our Goals
• Bridge AI and physics communities
• Contribute to multiple aspects of the EIC science, detector design, experimental control, simulation, analysis and theory
• Foster collaboration across labs, universities, and industry
• Engage, Collaborate, Innovate
• We invite you to share ideas, contribute to discussions, and help shape the future of EIC science by leveraging AI

📍 AI4EIC Workshop — MIT, Boston

📅 October 27–29, 2025

🌐 eic.ai

Backup

21