Define the physics project

Jet Physics with Machine Learning

Key words: jets, heavy flavour, PYTHIA, FASTJET, machine learning.

Basic set-up for event generation

(1) Set up a basic python code to generate events using PYTHIA event generator Generate simulated high-energy collision events using PYTHIA, including both events with heavy flavour jets and events without. Include the following conditions: CenterOfMass energy = 7 * TeV Collision system: particle1="proton", and particle2="proton" Process: "HardQCD:gg2ccbar = on" (charm-anti charm production)

(2) Store the events information (such as, pt, eta, phi, px, py, pz, E of the produced particles) in a file format (e.g., pickle, uproot etc).

Jet Clustering/ Reconstruction

(1) Read the uproot file using python code and retrieve the necessary information.

(2) Use a clustering algorithm (e.g. the anti-kT algorithm) to identify jets in the simulated events.

(3) Proceed for the jet reconstruction using "Fastjet" package. It involves few steps as follows: define the jet and provide "the name of algorithm", and "value of radius (R)" do clustering using "AwkwardClusterSequence". Provide the array of particles and jet def as arguments. transfer the reconstructed jets to a list.

(3) Find the jets with at least one D-meson and store the extract features such as, their energy (E), momentum (pt), mass (m), etc, into a file for further analysis. You could also use more sophisticated jet substructure techniques to extract additional information about the jets.

Jet Classification using Machine Learning

(1) Use machine learning to classify jets in simulated high-energy-collision events as containing heavy flavour (e.g. bottom or charm quarks) or not. This could potentially be useful in experimental analyses to identify and study these types of particles.

(2) Train a machine learning model (e.g. a neural network) to classify the jets as containing heavy flavour or not, using the extracted features as inputs. You could also try using more advanced machine learning techniques such as boosting or random forests.

(3) Evaluate the performance of your model on a test set of simulated events, and compare it to other classification methods (e.g. based on kinematic variables or more traditional methods like likelihood fits).

############################################################################################## This type of project could be a good opportunity to learn about both particle physics and machine-learning techniques, and could potentially lead to new insights into the properties of heavy-flavour particles in high-energy-collision events.

Jet Classification with CNN

Ref: LLNL-led-team-uses-machine-learning-derive-black-hole-motion-gravitational-waves

CNNs have been used successfully in a number of high-energy physics applications, including image classification in particle detectors and the reconstruction of particle tracks.

One possible approach for using CNNs to analyse heavy-flavour jets could be to train the CNN to classify jets based on their substructure, such as the presence of specific decay patterns or the presence of particular combinations of particles. This could potentially help identify jets originating from heavy-flavour particles with a higher level of accuracy than more traditional approaches.
Another possible application of CNNs in high-energy physics could be to improve the efficiency of event selection in particle detectors. For example, a CNN could be trained to identify rare signal events in a large dataset by recognizing characteristic patterns of particle interactions, which could help researchers find interesting events more efficiently.

open-physics / jet-physics-and-machine-learning