Choose timezone
Your profile timezone:
Powered by Indico
v3.3.5
Unlike the previous years, we will have a virtual round table this year. The anticipated format will be that there are 5-8min presentations introducing the audience to ongoing and recent projects related to ML on the DESY campus. The meeting is addressed to ML interested scientists and experts and is expected to last for ~2h. However, the motivation of the meeting is the same as in previous years, to get to know each other, and to be aware of the other projects and experts on campus, who use the same tools and algorithms, addressing entirely different scientific challenges.
Please register for this event to receive the most recent updates and connection details.
Start 10:30 AM
The meeting will be held as a zoom conference. To connect please use:
The M devision is pursuing an ambitious goal of bringing AI into accelerator operation within the next years. Several groups are involved in a broadening spectrum of AI and ML R&D activities, some of which I will present. This will include surrogate modelling and virtual diagnostics, reinforcement learning for accelerator operation, virtual environments for training and evaluation of control agents, and deep neural networks for modelling and control of accelerators.
The ML methods are explored to enable detecting crystals in SFX data and extract their properties online to avoid storing the entire dataset.
I will highlight some efforts at using Interpretable Machine Learning for HEP Pheno analysis and modeling COVID-19 data.
Accurate simulation of physical processes is crucial for the success of modern particle physics. However, simulating the development and interaction of particle showers with calorimeter detectors is a time consuming process and drives the computing needs of large experiments at the LHC and future colliders. Recently, generative machine learning models based on deep neural networks have shown promise in speeding up this task by several orders of magnitude. We investigate the use of a new architecture -- the Bounded Information Bottleneck Autoencoder -- for modelling electromagnetic showers in the central region of the Silicon-Tungsten calorimeter of the proposed International Large Detector. Combined with a novel second post-processing network, this approach achieves an accurate simulation of differential distributions including for the first time the shape of the minimum-ionizing-particle peak compared to a full GEANT4 simulation for a high-granularity calorimeter with 27k simulated channels. The results are validated by comparing to established architectures. Our results further strengthen the case of using generative networks for fast simulation and demonstrate that physically relevant differential distributions can be described with high accuracy.
CMS standard reconstruction algorithms are not designed for challenging signatures such as decays of long lived particles: we will show two applications that can overcome these limitations by using graph neural networks.
Bring your own cookies ;-)
I will give a brief overview of recent machine learning projects in our group. This will include graph networks as a promising tool for the reconstruction of complex data, statistical aspects of generative models, new tools for automated decorrelation and background estimation, as well as plans for hardware-accelerated ML.
DESY-IT, HZG and University Lübeck collaborate on the development of multimodal multitask learning methods involving the semantic segmentation of synchrotron CT data collected at Petra III.
We show the results of classification for two convolutional neural networks with different depth and architecture, by applying them to different representations of the same data from a single particle imaging (SPI) experiment.
We introduce a course of ML concepts for High Energy Physics-minded audience. The course balances between being concise, rich in theory and practical to build the skills required for modern HEP analysts. Infused with a classical Data Science approach, the course moreover includes variety of examples and an overview of recent ML applications in HEP. Aside from showcasing current developments we also discuss future prospects, plans and ideas.