Jan 29 – 31, 2020
University of Venda
Africa/Johannesburg timezone

An Alternative to Monte Carlo Generators with Deep Generative Models

Jan 30, 2020, 3:05 PM
University of Venda

University of Venda

Senate Chamber P/Bag X 5050 Thohoyandou, 0950


Mr Thabang Lebese (University of the Witwatersrand (ZA))


What is typically referred to as the inverse problem in High Energy Physics (HEP) can be described as the use of data to extract key information to build new a theory. The search of new resonances beyond the Standard Model (SM) involves the use of different Machine Learning techniques. For this purpose, based on the recent and major successes in the field of deep learning, particularly Deep Generative algorithms; Generative Adversarial Networks (GANs) which have been developed in less than a decade ago have proven to be of potential. The feasibility of addressing the inverse problem can be achieved via a combination of GANs and weak supervision. Weak supervision provides a way of combining the already known information about the backgrounds with the unknown hidden in the data, it is often used to extract features of the new Beyond the Standard Model signal from the data and with GANs used to create a Monte Carlo (MC) generator of the unknown signals with no significant loss in accuracy which could be better than classic MC.

Primary authors

Mr Thabang Lebese (University of the Witwatersrand (ZA)) Bruce Mellado Garcia (University of the Witwatersrand)

Presentation materials