RIST Popular Talk -53
Signal searches with deep-learning at the Large Hadron Collider
Date - 3rd March, 2022
RIST POPULAR TALK SERIES: 53
RESEARCH INSTITUTE OF SCIENCE AND TECHNOLOGY (RIST)
Schedule: Date: 3rd March, 2022.
Time: 2 pm.
Venue: RIST Office
Topic: Signal searches with deep-learning at the Large Hadron Collider: some phenomenological studies
By: Vishal Singh Ngairangbam
Theoretical Physics Division, Physical Research Laboratory,
Email: [email protected]
The Large Hadron Collider(LHC) is the most sophisticated machine built to probe physics at the sub-nuclear length scales. It generates huge amounts of data that are stored all over the world in the worldwide LHC computing grid. Therefore, the physics goals of the LHC present in itself a humongous data analysis task. The particle physics community has been using machine learning techniques in data analysis for decades.
However, with the advent of modern deep-learning algorithms propelled by the wide availability of high-end GPU acceleration and a rich research ecosystem unearthing state-of-the-art algorithms and architectures, there is an unprecedented increase in applying such algorithms at various stages of the analysis pipeline.
In this talk, the speaker will discuss the improvement in the searches of invisible Higgs decays produced via Vector Boson Fusion (VBF) with Convolutional Neural Networks (CNNs). The analysis of such signals relies on accurate physics simulations at different energy scales extensively validated on experimental data.
Deep-learning algorithms can utilize minute differences in the data; therefore, it is imperative to scrutinize the phenomenological implications of various aspects in the simulation. For the VBF production, we also study the dependence of the CNN's performance on the recoil scheme used in the parton shower and the perturbative accuracy of the matrix-element simulation of the hard process and find that the performance varies for different signal simulations.
The result above indicates that neural networks, although highly expressive, are not well understood from a physical perspective. As a step towards resolving these issues, I'll discuss "Energy-weighted Message-Passing," an infra-red and collinear safe graph neural network algorithm with more attractive phenomenological properties.
N. Nimai Singh,
Convener, RIST popular Talk Series
Research Institute Of Science And Technology (RIST)
Manipur University Complex,
Imphal – 795 001, India
* This information is sent by N. Nimai Singh who can be contacted at nimai03(AT)yahoo(DOT)com
This Post is webcasted on March 01 2022
* Comments posted by users in this discussion thread and other parts of this site are opinions of the individuals posting them (whose user ID is displayed alongside) and not the views of e-pao.net. We strongly recommend that users exercise responsibility, sensitivity and caution over language while writing your opinions which will be seen and read by other users. Please read a complete Guideline on using comments on this website.