IndoFrench High Energy Physics Collaboration

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
==== Jetology Working Group (includes substructure,  heavy objects, top physics, boosted stuff,…) ====
 
Tuhin S. Roy (Conv.), Santosh K. Rai, Dilip Kumar Ghosh, Abhishek Iyer, Fawzi Boudjema, Aldo Deandrea, Rohini Godbole

===2018 Progress Report===

== Project 1: Leptonic Top Tagger using Fat Jets ==
Suman Chatterjee, Rohini Godbole, Tuhin S. Roy, and Seema Sharma

In this work, we perform an object tagging approach to identify a boosted top quark decaying leptonically, for each lepton flavor (electron, muon and tau).  We start from a fat jet and construct several observables (various substructure variables using information from different parts of the detector), but refrain from using any global event information like missing energy. In this course we also develop some new observables which can act as discriminators for the leptonically decaying top quark to non-SM background sources, like supersymmetric partner of top whose decay can result in an identical final states of observed particles (i.e. lepton inside a b-jet).
  
{{ :wiki:leptop.jpg?750}}

The figures depict the performance of the tagger for top decaying to electron, muon, tau (which further decays leptonically)  respectively as the signal with respect to backgrounds (b-jets as well as q/g jets).


== Project 2: An Autoencoder for Anomaly Finder ==
Vijay H. Aravind and Tuhin S. Roy

An autoencoder is an architecture of artificial neural network which can be trained in an unsupervised way to compress the input into an efficient and small intermediate representation which can then be used to obtain the input back. In jet substructure analysis, we can use this technique to **veto** QCD jets in an event. The main advantage of this procedure (as compared to the traditional classification methods) is that since this technique is unsupervised, it only requires QCD jets to train. After converting a jet into an image (basically a 2D histogram of energy or pt of the jet in the eta - phi plane), we pass it through a novel preprocessing step that takes into account the extra freedom in Lorentz Boosting. An autoencoder is trained afterwards to learn about these processes QCD jets.    

{{:wiki:autoencoderroc.jpg?200|}}

To quantify the performance of the anomaly (anything but QCD-jets) finder we show the ROC plots where hadronically decaying boosted top jets are injected as signals. For comparisons, we also show the performance of HEPTopTagger, which uses all information of top but does a substructure based tagging, and results of a recently proposed tagger based on autoencoder (1808.08992).


== Project 3: Tagging di-tau jets ==
Vijay H. Aravind and Tuhin S. Roy

The main objective of this work is to devise a tagger for a jet consisting of decay products of two hadronically decaying tau particles (i.e. di-tau jets). In order to do this, we are constructing two alternative taggers, one based on jet-substructure variables (such as NSubjettiness, number of hard tracks, planer flow etc.) and the other entirely based on deep convolutional neural networks.
IndoFrench High Energy Physics Collaboration

User Tools

Site Tools

Page Tools
