Track reconstruction at LHC as a collaborative data challenge use case with RAMP

Abstract

Abstract. Charged particle track reconstruction is a major component of data-processing in high-energy physics experiments such as those at the Large Hadron Collider (LHC), and is foreseen to become more and more challenging with higher collision rates. A simplified two-dimensional version of the track reconstruction problem is set up on a collaborative platform, RAMP, in order for the developers to prototype and test new ideas. A small-scale competition was held during the Connecting The Dots / Intelligent Trackers 2017 (CTDWIT 2017) workshop. Despite the short time scale, a number of different approaches have been developed and compared along a single score metric, which was kept generic enough to accommodate a summarized performance in terms of both efficiency and fake rates. Data challenges and RAMPAdvances in machine-learning technology is moving towards a new paradigm in data processing design, in which the role of the individual will have to be redefined. In order to integrate the human creativity with the computational resources in an efficient way, various data-challenge platforms have been emerging, which provide a competitive and/or collaborative environment to crowd-source the analysis expertise.

The Tracking Machine Learning Challenge: Accuracy Phase

“…The TrackML challenge has been a long running competition series to gather new algorithmic ideas to speed up tracking in the LHC experiments. After the first round of initial discussions, a prototype challenge [2] was organised during the Connecting The Dots workshop (an annual workshop for experts in pattern recognition) held at IJCLab in Orsay in March 2017. The problem was essentially the same as the one exposed here but significantly simplified to be a 2D problem with just 20 tracks per event (instead of 10.000 in 3D).…”

“…The Tracking Machine Learning (TrackML) challenge took place in two phases, an Accuracy phase [1] in 2018 on the Kaggle platform , and a Throughput phase in 2018-2019 on Codalab , preceded by a limited scope 2D prototype competition [2]. This paper is documenting in details the Throughput phase, which combined accuracy and inference speed, while only the minimal summary of the Accuracy phase is given (see [1] for details).…”

“…Recent attempts of applying Machine Learning to particle physics pattern recognition-tracking indicate a strong potential [20]. A one day hackathon [5] limited to a twodimensional problem has shown the richness of approaches and the setup to be tested.…”

Design optimization of a pixel-based range telescope for proton computed tomography

“…Several strategies for the tracking algorithm have been explored, with the main idea being a "track-following" scheme [21]. Some of the ideas behind the algorithm, such as calculating a weight for each hit, based on the angular change of the track, S , and comparing the accumulated weight to against a global maximum, S max , are described in [16] as the hyperbelle_tree_6 solution to a pre-defined tracking challenge. In [9,13] an earlier version of this algorithm was briefly described, however there it was only applied on a broad and uniform irradiation field, and with inferior performance.…”

“…a higher incident beam intensity. In this study, such a tracking algorithm is proposed, based on similar experiments (mainly in High Energy Physics [16]) and prior experience with the DTC [9,13].…”