Jesse Ernst
PhD University of Rochester (1995)
Research Associate University of Illinois, Urbana-Champaign (1995-1998)
Research Assistant Professor University of Illinois, Urbana-Champaign (1998-2001)
Research
Research Areas:
Elementary Particle Physics, Information Theory, Machine Learning
Current Research:
My current research focus lies at the intersection of Machine Learning (ML), Information Theory, and Particle Physics. In one area of work, we use information-theoretic tools to address a significant shortcoming in using ML models for classification in particle physics: there are few reliable ways to quantify how much information is present in the measured observables. We have shown that the Mutual Information between a set of measured variables and the event classes can provide this. If one can compute it accurately, it gives a fast, algorithm-independent benchmark that sets an upper limit on how well any classifier can distinguish between classes. Because this benchmark can be computed far more quickly than training a full ML model, it has an important application in feature selection. With it, one can quickly compare potential sets of discriminating variables without ever building a classifier.
Selected publications related to current focus:
N. Carrara and J. Ernst, "Using Monte Carlo Tree Search to Calculate Mutual Information in High Dimensions," [arXiv:2309.08516 [physics.data-an]]. (Sep 2023)
D. S. Akerib et al. [LUX], "Fast and flexible analysis of direct dark matter search data with machine learning," Phys. Rev. D 106, no.7, 072009 (2022) doi:10.1103/PhysRevD.106.072009 [arXiv:2201.05734 [astro-ph.CO]].
N. Carrara and J. Ernst, "On the Estimation of Mutual Information," (2019) [arXiv:1910.00365 [physics.data-an]].
N. Carrara and J. A. Ernst, "On the Upper Limit of Separability," (2017) [arXiv:1708.09449 [hep-ex]].
Collaboration memberships and related publications:
I am a long-term continuing member of the BaBar Collaboration and a former member of the ATLAS and CLEO Collaborations