Focus on:  Hide Contributions
Login

2nd Round Table on Machine and Deep Learning at DESY

→ Europe/Berlin
Seminar Room (Bldg. 28C Flash Hall)

Seminar Room

Bldg. 28C Flash Hall
Description

The number of activities on machine and deep learning within the heterogenous research landscape on DESY campus is continuing to rise. To enable and strengthen inter-department comm-unication and collaboration between the ML/DL experts on campus we invite you to the 2nd Round Table on Machine and Deep Learning at DESY.

If you are already working with or plan to work machine or deep learning, you are welcome.

Please register for this event (primarily to ensure we have enough coffee).

    • 10:00 → 10:10
      Welcome
    • 10:10 → 11:30
      ML/DL @ DESY-campus 1

      Presentations from DESY campus colleagues on ML/DL

      • 10:10
        Convolutional Neural Networks for electron microscopy tomography 20m
        Finding instances of a protein transmembrane complex using segmentation and classification with CNNs is presented. The presented project is a collaboration between DESY-IT and CSSB
        Speaker: Dr Philipp Heuser (DESY)
        Slides
      • 10:30
        Generative Models for Fast Calorimeter Simulation 20m
        Speaker: Engin Erin (DESY)
        Slides
      • 10:50
        ML methods for FEL scattering data analysis 20m
        Speaker: Dr Alireza Sadri (DESY)
      • 11:10
        Machine learning in the DESY ATLAS group 20m
        Speaker: Christopher Pollard (DESY)
        Slides
    • 11:30 → 11:45
      Coffee Break
    • 11:45 → 13:05
      ML/DL @ DESY-campus 2
      • 11:45
        Automation of CMS workflow recovery 20m
        Attempts to use ML to predict the operator action on the CMS simulation and data processing failed jobs.
        Speaker: Dr Hamed Bakhshiansohi (DESY)
        Slides
      • 12:05
        Neural networks for small angle scattering data analysis. 20m
        We propose a novel method of SAXS data analysis based on application of interconnected neural networks (perceptrons). For given experimental data from proteins, RNA or DNA our stack of networks evaluates Rg, Dmax, MW, p(r) and a noise-free scattering curve. This completely automatic approach has proved to be robust against experimental errors, applicable to data from particles of various nature, size, and shape. The method was implemented as a publicly available web service with a graphical interface, providing the possibility to inspect and download the results (https://dara.embl-hamburg.de/gnnom.php).
        Speaker: Dr Dmitry Molodenskiy (EMBL)
        Slides
      • 12:25
        Anomaly Detection for SRF Cavities 20m
        The European XFEL uses superconducting radiofrequency cavities for the electron acceleration. In total 808 cavities are operated in pulsed mode with a 10 Hz repetition rate, which amounts to ~700 Million RF-pulses in 24 hours. This talk explores the possibility of monitoring anomalous behavior of individual cavities and determining levels of anomaly for each pulse.
        Speaker: Ayla Nawaz (DESY/MSK)
      • 12:45
        Classification for Single Particle Imaging experiments 20m
        Speaker: Dr Alexandr Ignatenko (DESY)
    • 13:05 → 14:00
      Lunchbreak

      coffee, refreshments and open discussion

    • 14:00 → 15:00
      Snap ML - Accelerated Machine Learning for Business

      Haris Pozidis, PhD
      Manager Cloud Storage & Analytics, Principal RSM, Master Inventor, IBM Research

      Snap Machine Learning (Snap ML) is a new software library for training machine learning models that is characterized by very high performance, scalability to TB-scale datasets and high resource efficiency. It continuously evolves and currently supports generalized linear models, decision trees, random forests and gradient boosting machines. Snap ML has been built to address the needs of business applications, which often have to deal with data that is big in size, react fast to changing environments, and use resources efficiently to drive down cost. In addition, with the most recent addition of a new gradient boosting model, Snap ML offers high generalization accuracy, which drives higher profits in AI-infused business applications. In this talk I will present the principles of the Snap ML library, explain how it achieves high speed and scalability, and present several cases of business workloads that demonstrate the benefits reaped by Snap ML.

      slides
    • 15:00 → 16:00
      Discussion and Conclusions
      Convener: Frank Schluenzen (DESY)
      slides
Powered by Indico v3.3.5