Machine Learning and Science Forum — First-class machine learning model for Science: Graph Neural Network

ML&Sci Forum

November 9, 2020
11:00am to 12:00pm
Virtual Participation

Machine Learning and Science Forum 
Date: Monday, November 9, 2020
Time: 11:00 AM - 12:00 PM Pacific Time
Location: Participate remotely using this Zoom link 

First-class machine learning model for Science: Graph Neural Network

Xiangyang Ju, Lawrence Berkeley National Laboratory

This talk will dive deep into the graph neural network for science with a biased focus on exemplary applications from High Energy Physics. Data collected from the detectors at the Large Hadron Collider (LHC) are usually dynamic in sizes, sparse in density and high dimensions in representations. Such data cannot be fully represented by feature vectors, event images or point clouds without scarifying the information loss. However, graphs can naturally represent such science data because of its unlimited expressive power. In addition, the fast advancing geometrical learning generalizes convolutional and recurrent neural networks to datasets with arbitrary geometry and sparsity. It boosts the exploration of deeper hidden graph attributes, the learning of the global and local relational information in the graph. It can even form parameterized message-passing through which information is propagated across the graph, ultimately learning sophisticated graph attributes. The GNN architecture is unique, so are its computational characteristics.

The talk will use the workflow developed by the Exa.TrkX collaboration for reconstructing charged particles' trajectories for the High-Luminosity LHC as an example to demonstrate the power of GNN. Other HEP projects based on GNN will be briefly discussed as well. In addition, the talk will profile GNN's computational features and discuss the implication in terms of real-time data processing. The talk ends with an outlook of GNN for HEP and for science.

The BIDS Machine Learning and Science Forum (formerly the Berkeley Statistics and Machine Learning Forum) was launched in Spring 2018 and currently meets biweekly (during the spring and fall semesters) to discuss current applications across a wide variety of research domains in the physical sciences and beyond. Hosted by BIDS Faculty Affiliate Uroš Seljak (professor of Physics at UC Berkeley) and BIDS Research Affiliate Ben Nachman (Physicist and Staff Scientist at Lawrence Berkeley National Laboratory), these active sessions bring together domain scientists, statisticians, and computer scientists who are either developing state-of-the-art methods or are interested in applying these methods in their research. All interested members of the UC Berkeley and Berkeley Lab communities are welcome and encouraged to attend.  To receive email notifications about upcoming meetings, or to request more information, please contact berkeleymlforum@gmail.com.

Speaker(s)

Xiangyang Ju

Physics Division, Lawrence Berkeley National Laboratory