Alexei Tam

Advisor: Richard Cavanaugh

Mentor: Kevin Pedro

Undergraduate: University of Illinois Chicago (Physics)

Graduate: University of Illinois Chicago (Physics)

Project: hackjet - Jet Clustering with SIFT (Scale-Invariant Filtered Tree) and hierarchical agglomerative clustering

Focused on improving the computational efficiency of jet clustering that will be used in searches for new phenomena at the HL-LHC. Historically, a wide variety of infrared and collinear safe jet clustering algorithms have been used, such as: the family of kT algorithms, JADE, DURHAM, etc. However, many of these algorithms, namely those of the kT-family, rely on an R_0 parameter that, while granting a variable configurability to jet sizes, relies on some understanding of the Parton shower process being studied, which becomes problematic for trying to understand BSM and dark sector processes. One algorithm, SIFT (Scale-Invariant Filtered Tree), successfully provides a scale-invariant measure that does not depend on such parameters. However, the SIFT metric is complicated and does not satisfy the triangle inequality in Euclidean geometry, so tools that are used to speed up kT-family like FastJet cannot assist SIFT. Hierarchical agglomerative clustering algorithms can be used, but the complexity of such algorithms is O(n^3). Alexei is studying a variant known as heap-based agglomerative clustering algorithm, that holds a promising solution forward as it is expected to have O(n^2 log(n)) complexity. Such algorithms may have the potential to improve the speed of the kT-family of algorithms also. So far, a survey of literature has been conducted and a baseline (naive) O(n^3) SIFT algorithm is currently being implemented. The ultimate goal for this project is to implement the SIFT jet clustering algorithm on GPUs in anticipation of the HL-LHC.