URL: https://physikseminar.desy.de/zeuthen/past_colloquia/colloquia_in_2017/february_15_2017/@@siteview
Breadcrumb Navigation
Study Machine Learning through HEP
Andrey Ustyuzhanin (Higher School of Economics, National Research University, Moscow | Yandex Data Factory)
Seminar room 3, 15:00
Applications of advanced machine learning (ML) techniques become more and more widespread in everyday life. Natural sciences, being one of the best places to test new approaches, benefit greatly from this development. In my talk I will briefly cover several problems that are characteristic for the High-Energy Physics: trigger systems optimization, and data certification pipeline automation. Then I’ll show how those challenges can be addressed from Machine Learning perspective. Examples of advanced applications of traditional ML approaches to offline data analysis will be given. Also I will cover some tools that allow for significant simplification of data analysis code sharing and collaboration.
The solutions to be described have been developed by Yandex research team in collaboration with LHCb and are used as materials for training course on practical applications of Machine Learning at Yandex School of Data Analysis.
This picture is constructed as LHCb event display that has passed through popular Deep Learning algorithm - Style Transfer - that borrows the style from Van Gogh’s ‘Starry Night’ (https://github.com/jcjohnson/neural-style). So it is somewhat metaphorically related to the nature of the talk (learning algorithms for HEP problems borrowed from other domains).
