CERN Accelerating science

Yandex, Russia’s leading search engine, joined CERN openlab as an Associate in early 2013 to provide computing resources and proprietary data-processing technologies — including the machine-learning technology MatrixNet — to CERN researchers. MatrixNet was already used for analysing B-meson decay data at the LHCb experiment, one of the four big LHC detectors. While this was just the first case of the technology’s application at CERN, it encouraged Yandex to provide MatrixNet for analysis of a variety of other decays and further complex physical processes. The technology enables physicists to perform filtering of huge datasets and find extremely rare events.

During 2013, Yandex provided 5.2% of the total worldwide computing resources available to the LHCb experiment, making it the seventh largest provider of such resources, after the LHCb High-Level Trigger farm; CERN; and the Italian, French, German and British national computer centres for High Energy Physics.

Yandex used its search engine technology to provide a prototype index of LHCb events. The large LHCb datasets are typically analysed sequentially to select events satisfying specific criteria. Selecting particular types of events using traditional methods can take several hours on large computer farms. The index developed by Yandex enables LHCb physicists to find within seconds the rare events of interest for their analyses. Geared specifically to LHCb needs, this search system saves both time and resources and can greatly speed up the analysis.

In a second project, Yandex applied its MatrixNet technology as an alternative method of selecting interesting events from the LHCb datasets. By comparing the Yandex software with the algorithms that were used in the past, LHCb physicists managed to significantly improve their science output. The improved understanding of the difference between these old and new methods helped the physicists refine both techniques.

Previous activities for the Project with Associate covering 2012 are available here.