Revision as of 14:06, 11 February 2025

Purpose

The overall topic is: „Best practice for AI in nuclear physics applications“. We wish to cover the following items via plenary talks and/or interactive tutorials:

• Feature engineering --> Feature normalization, correlation coefficients, feature selection, etc.
• Overfitting --> Dropout layers, weight regularization, etc.
• Performance evaluation --> ROC-Curve, confusion matrix, accuracy, loss curves,...
• Visualization --> How to properly present the performance of a model ? What are "good" diagnostic plots ?
• Model deployment --> How to use a model within the GlueX analysis software
• Data fed into models --> What data sets are used ? Numerical, vs. Images, raw data vs. clean data,...
• Tools made available by the data science department
• Optional, depending on time: HPO --> Tune the parameters of your model
• Optional, depending on time: Bookkeeping of models via MLFlow

Location and Time

The workshop will take place at:

DATES: February 18 (all day) - 19 (morning only), 2025

LOCATION: CEBAF Center F113

Remote Participation

References

Workshop Software

PID example Notebook

Agenda

AI Tutorials

Feb 18

• 09:00 Introduction (15min)
• 09:15 Notebook setup (15 min)
• 09:30 Data preparation and feature engineering (1h)
• 11:00 Setting up a model and training it (1h)

• 13:30 Final performance evaluation (1h)
• 14:30 HYDRA ()

Feb 19

Registration

Please add your name to the list of attendees below. No formal registration or registration fee is required.

Recording