Powered by Indico
v3.2.9
This is the second event of a series on Machine Learning applications in particle physics. The afternoon will be dedicated to "lightning talks" that overview the applications of machine learning currently being exploited in INFN-related activities in Rome. The goal is to share with the rest of the community the challenges the various groups are trying to tackle with machine learning, and the techniques they are using (and why).
The agenda of the previous, introductory event on state-of-the-art ML techniques and their applications in high-energy physics may be found on https://agenda.infn.it/event/33944/
The event will be in presence. A Zoom connection will also be available for those unable to join.
Valerio Ippolito
CYGNO is a directional Dark Matter detector with optical readout at LNGS, Italy. CYGNO uses a camera with scientific CMOS (sCMOS) sensor for readout and it produces high resolution images which are well suited for Machine Learning Analysis (using Convolution Neural Networks) because of their high granularity and low noise. Additionally, accurate Monte Carlo simulations of the tracks are produced and these simulated tracks are analysed using classical Convolution Neural Networks to classify them into different classes such as Electron and Nuclear recoil.
Track finding in high-density environments is a key challenge for experiments at modern accelerators. In this presentation we describe the performance obtained running machine learning models studied for tracking in the ATLAS Muon High Level Trigger. These models are designed for hit position reconstruction and track pattern recognition with a tracking detector, on a commercially available Xilinx Alveo U50 and Alveo U250.
The design flow of an online ML-based partial particle identification system for the NA62 RICH detector will be presented and discussed.
We present the developed methodologies and the results of the implementation in a working engineering pipeline for an ultra-light and ultra-fast Convolutional Neural Network for muon identification in the future L0 Muon Trigger of the ATLAS experiment at HL-LHC.
Despite decades of theoretical studies, the nature of the glass transition remains elusive and debated, while the existence of structural predictors of its dynamics is a major open question. Recent approaches propose inferring predictors from a variety of human-defined features using machine learning. Here we determine the long-time evolution of a glassy system solely from the initial particle positions and without any handcrafted features, using graph neural networks as a powerful model.
We study the possibility to implement HLT trigger filtering algorithms based on deep neural network, and to use commercial accelerators boards based on FPGA processors to improve the trigger performance in terms of processing time per event and processed event throughput. We show a proof of concept based the interesting use-case of the identification of long-lived particles in the muon spectrometer of the ATLAS experiment.
We discuss how complicated likelihoods obtained from phenomenological fits can be reproduced with suitably trained DNNs, with a dramatic increase in speed. We highlight a few ideas on how this approach can be extended in several directions.
We propose a biologically grounded multi-area thalamo-cortical plastic spiking neural network model and investigate the role of NREM - REM cycles on its awake cognitive performance. We demonstrate that sleep has a positive effect on energy consumption and cognitive performance during the post-sleep awake classification task of handwritten digits. NREM and REM simulated dynamics modify the synaptic structure into a sharper representation of training experiences. Sleep-induced synaptic modifications reduce firing rates and synaptic activity without reducing cognitive performance. Also, it creates novel multi-area associations. The model leverages the apical amplification, isolation and drive experimentally grounded principles and the combination of contextual and perceptual information. In summary, the main novelty is the proposal of a multi-area plastic model that also expresses REM and integrates information during a plastic dream-like state, with cognitive and energetic benefits during post-sleep awake classification.
Efficient and accurate reconstruction and identification of of particle decays plays a crucial role in the program of measurements and searches under study for the future high-energy particle colliders. Leveraging recent advances in Machine Learning algorithms, that have dramatically improved the state-of-the-art in visual object recognition, we have developed novel tau identification methods able to classify tau decays in leptons and hadrons, and to discriminate them against QCD jets. We present the methodology and the results of the application at the interesting use case of the IDEA dual-readout calorimeter detector concept proposed for the future FCC-ee electron-positron collider.
We aim to implement a methodology that helps archaeologists' reconstruction work of fragmented 3D artefacts(objects). Deep Learning networks make it possible to tackle problems whose complexity prevents the use of more classical techniques.
A tracking system is under development within the “SIREN” project, with the aim of collecting useful data to reconstruct the dynamics of abnormal events in Nuclear Medicine Therapy (NMT), with potential risk of exposure for workers.
This device will provide the real-time tracking of the body pose of workers inside a NMT room, to estimate the radiation source-operator distance and direction during the occurrence of the event. It relies on the video streams of two high resolution cameras recording the same scene from two viewpoints, and a Human Pose Estimation (HPE) algorithm based on a convolutional neural network model that can detect up to 17 joints in the human body. The developed tracking system has proven to be simple to set up, cost-effective and adequate in terms of distance accuracy and computation time when applied to the spatial dimensions of NMT rooms. The tracking data can also provide valuable insights into the assessment of the dose to operators, by integrating their tracked positions within computed radiation field maps.
The Virgo group in Rome uses different Deep Learning algorithms to search for gravitational wave signals. We will illustrate the techniques currently employed and ongoing research.
Design of an experimental physics equipment, which is essentially assimilated to the optimization of multi-parametric systems, can considerably benefit of AI-inspired computational approaches. A quick overview of some of the current applications in AI-based methods for the definition and design of spectrometer and accelerator components will be presented, driven by the ongoing efforts within the Electron-Ion-Collider (EIC) project.
Developing and testing methodologies that allow to interpret the predictions of the AI algorithms in terms of transparency, interpretability, and explainability has become today one of the most important open questions in AI. We will present the activity of the MUCCA project, funded by CHIST-era (Multi-disciplinary Use Cases for Convergent new Approaches to AI explainability). Both Sapienza and INFN are partners in this project. We will focus on the Rome group use cases (medical imaging and neuroscience).
We investigate the possibility to apply quantum machine learning techniques for data analysis, with particular regard to an interesting use-case in high energy physics. We propose an anomaly detection algorithm based on a parametrized quantum circuit. This algorithm has been 3 trained on a classical computer and tested with simulations as well as on real quantum hardware. The quantum anomaly detection algorithm is able to detect complex structures like anomalous patterns in the particle detectors produced by the decay products of long-lived particles produced at a collider experiment.