→
Europe/Zurich
IdeaSquare (CERN)
IdeaSquare
CERN
Christopher Edward Brown
(CERN),
Ioannis Xiotidis
(CERN),
Maciej Mikolaj Glowacki
(CERN),
Rimsky Alejandro Rojas Caballero
(CERN)
Description
This workshop is an intiative of the Next Generation Triggers project and aims to bring together intra- and inter-experiment groups to start discussions on the effective deployment of Machine Learning algorithms into hardware trigger systems and beyond in anticipation of the HL-LHC.
The day will be divided into two sessions: the morning will consist of ATLAS and CMS working groups, and will be focused on discussing experiment-specific requirements, while the afternoon will be an open discussion, organised around three common themes in ML deployment:
- Training: Strategies for training models deployed under evolving conditions
- Model Logistics: Managing models through software, firmware, and software transitions
- Data Flow: Handling data conversions with distributed, heterogeneous computing
There will be short talks at the start of each theme.
The deadline for registration is May 23th, 2025.
Contact
Participants
-
-
Welcome Session - Practical Information
-
MLOps in ATLAS
-
1
Firmware: ML integration into AthenaSpeaker: Kenny Jia (Stanford University/SLAC)
-
2
Discussion: Bitwise ML simulationSpeaker: Ioannis Xiotidis (CERN)
- 3
-
4
Discussion: Pythonisation of training filesSpeaker: Ioannis Xiotidis (CERN)
-
10:45
Coffee Break
-
5
Hardware: Infrastructure for developing MLSpeaker: Rimsky Alejandro Rojas Caballero (CERN)
-
6
Discussion: Hardware for ML deploymentSpeaker: Ioannis Xiotidis (CERN)
-
1
-
MLOps in CMS
-
7
Software: Model matadata and Cond DB interfaceSpeaker: Davide Valsecchi (ETH Zurich (CH))
-
8
Firmware: Reconfigurable FW and monitoringSpeaker: Christopher Edward Brown (CERN)
-
10:45
Break
-
9
Hardware: Infrastructure for ML deploymentSpeakers: Hannes Jakob Hansen, Maciej Mikolaj Glowacki (CERN), Raulian-Ionut Chiorescu, Ricardo Rocha (CERN)
-
7
-
Lunch Break
-
Open Session - Discussion
-
10
Training: Strategies for training models deployed under evolving detector conditionsSpeaker: Christopher Edward Brown (CERN)
-
11
Open Floor: Discussion & Questions
-
12
Model Logistics: Managing models through software, firmware, and back to software againSpeaker: Ioannis Xiotidis (CERN)
-
13
Open Floor: Discussion & Questions
-
14
Data Flow: Handling data conversions and distributed, heterogeneous computing environmentsSpeakers: Leon Joel Kerner (CERN | Rheinisch Westfaelische Tech. Hoch. (DE)), Maciej Mikolaj Glowacki (CERN)
-
15
Open Floor: Discussion & Questions
-
10
-
Workshop PictureConvener: Alex Lasa Lamarca
-