BNL AI/ML Working Group Seminar Series

US/Eastern
Description

Seminar: Monthly seminar on recent research in AI/ML. 

    • 12:00 13:00
      Keeping the human in the loop: autopilot for experiments at diffraction beamlines 1h

      References:
      https://www.nature.com/articles/s43588-021-00059-2
      https://www.nature.com/articles/s41524-021-00575-9
      https://iopscience.iop.org/article/10.1088/2632-2153/abc9fc
      https://arxiv.org/abs/2104.00864

      Speaker: Phillip Maffettone
    • 12:00 13:00
      Provably exact sampling for first-principles theoretical physics 1h

      Abstract: In the context of lattice quantum field theory calculations in particle and nuclear physics, I will describe avenues to accelerate sampling from known probability distributions using machine learning. I will focus in particular on flow-based generative models, and describe how guarantees of exactness and the incorporation of complex symmetries (e.g., gauge symmetry) into model architectures can be achieved. I will show the results of proof-of-principle studies that demonstrate that sampling from generative models can be orders of magnitude more efficient than traditional Hamiltonian/hybrid Monte Carlo approaches in this context.

      Biography: Phiala Shanahan obtained her BSc and PhD from the University of Adelaide in 2012 and 2015. Before joining the MIT physics faculty in 2018, Prof. Shanahan was a Postdoctoral Associate at MIT and held a joint position as Assistant Professor at the College of William & Mary and Senior Staff Scientist at the Thomas Jefferson National Accelerator Facility. Prof. Shanahan is the recipient of Early-Career Awards from both the NSF and the DOE, and in 2021 was awarded the Maria Goeppert Mayer Award of the American Physical Society.

      Speaker: Phiala Shanahan (MIT)
    • 12:00 13:00
      High Performance FPGA Embedded System for Machine Learning Based Tracking and Trigger in sPhenix and EIC 1h

      We present a comprehensive end-to-end pipeline to classify triggers versus background events in this paper. This pipeline makes online decisions to select signal data and enables the intelligent trigger system for efficient data collection in the Data Acquisition System (DAQ) of the upcoming sPHENIX and future EIC (Electron-Ion Collider) experiments. Starting from the coordinates of pixel hits that are lightened by passing particles in the detector, the pipeline applies three-stages of event processing (hits clustering, track reconstruction, and trigger detection) and labels all processed events with the binary tag of trigger versus background events. The pipeline consists of deterministic algorithms such as clustering pixels to reduce event size, tracking reconstruction to predict candidate edges, and advanced graph neural network-based models for recognizing the entire jet pattern. In particular, we apply the Message-Passing Graph Neural Network to predict links between hits and reconstruct tracks and a hierarchical pooling algorithm (DiffPool) to make the graph-level trigger detection. We obtain an impressive performance (>= 70% accuracy) for trigger detection with only 3200 neuron weights in the end-to-end pipeline. We deploy the end-to-end pipeline into a field-programmable gate array (FPGA) and accelerate the three stages with speedup factors of 1152, 280, and 21, respectively.

      Biography: Dantong Yu received a BS degree in computer science from Peking University and a Ph.D. degree in Computer Science from University at Buffalo. He joined Martin Tuchman School of Management at the New Jersey Institute of Technology in 2016. Before that, he worked for Brookhaven National Laboratory for sixteen years and collaborated with Physicists on multiple experiments, including ATLAS, STAR, and PHENIX. He founded and led the Computer Science Group in BNL between 2009 and 2016. His research interests include data mining, machine learning, high-speed network, and communication protocols. He has published 70 papers in leading technical journals and conferences. He has served on the review panels for NSF, DOE Early Career Investigator, and DOE SBIR/STTR. He is a PC member of KDD, ICDM, CIKM, ICDE, ICCCN, HiPC, and ICPADS.

      Speaker: Dantong Yu (New Jersey Institute of Technology)
    • 12:00 13:00
      When reinforcement learning meets quantum computing 1h

      Recently, reinforcement learning (RL) has demonstrated various applications with superhuman performance such as mastering the game of Go. Meanwhile, the development of quantum computing hardware shed light on building practical quantum applications to tackle previously unsolved problems. What will happen if we combine these two fascinating techniques? In this talk, I will present the recent progress in quantum RL as well as using classical RL to help certain tasks in quantum computing.

      Speaker: Yen-Chi (Sam) Chen (Brookhaven National Laboratory)