Self-Supervised Learning for End-to-End Particle Reconstruction for the CMS Experiment

Description

One of the important aspects of searches for new physics at the Large Hadron Collider (LHC) involves the identification and reconstruction of single particles, jets and event topologies of interest in collision events. The End-to-End Deep Learning (E2E) project in the CMS experiment focuses on the development of these reconstruction and identification tasks with innovative deep learning approaches.

Transfer learning and Self-Supervised learning techniques often outperform traditional models that are trained from scratch. State-of-the-art SSL (Self-Supervised Learning) techniques, when applied to unlabelled data, learn useful features without annotation. Later, when these models are finetuned with the limited labelled data, a higher level of generality is observed. This project explores Self-Supervised Contrastive/Clustering/Distillation based Learning techniques to build robust models that perform better than the existing supervised baselines.

Duration

Total project length: 175/350 hours.

Task ideas

Explore Contrastive/Clustering/Distillation based techniques to train an encoder with rich feature vectors,
Extend the study by fine-tuning with labelled data and comparing the results with pre-existing supervised baselines.
Extend the idea of SSL to hybrid (Conv + ViT) architectures and compare scores for downstream tasks.

Expected results

Development of a robust SSL-based pipeline that can be used to train neural networks which can be later fine-tuned to produce state-of-the-art results for downstream tasks.

Difficulty level

Advanced

Requirements

C++, Python, PyTorch, Tensorflow and some previous experience in Deep Learning.

Test

Please use this link to access the test for this project.

Mentors

Emanuele Usai (University of Alabama)
Diptarko Choudhury (NISER)
Sergei Gleyzer (University of Alabama)
Ruchi Chudasama (University of Alabama)
Eric Reinhardt (University of Alabama)
Colin Crovella (University of Alabama)

Please DO NOT contact mentors directly by email. Instead, please email ml4-sci@cern.ch with Project Title and include your CV and test results. The mentors will then get in touch with you.