Physics Group Meeting
Information Geometry and Machine Learning as Tools for Phenomenology
I will discuss particle physics phenomenology from the point of view of statistical inference.
I will describe how the concept of information geometry can be used to equip the parameter space of a theory with a metric.
The central object in this construction is a likelihood function, which is typically intractable once non-perturbative and experimental effects are incorporated.
I will then describe how machine learning can be used to effectively approximate these intractable likelihoods in realistic settings. If time allows I will discuss extensions of these ideas in the context of jet physics and lattice QCD.
Suggested Reading:Better Higgs boson measurements through information geometry arXiv:1612.05261Constraining Effective Field Theories with Machine Learning arXiv:1805.00013 (short PRL)arXiv:1805.00020 (long PRD)QCD-Aware Recursive Neural Networks for Jet Physicshttps://arxiv.org/pdf/1702.00748.pdfJUNIPR: a Framework for Unsupervised Machine Learning in Particle Physicshttps://arxiv.org/pdf/1804.09720.pdfMachine learning action parameters in lattice quantum chromodynamicshttps://arxiv.org/abs/1801.05784
I will describe how the concept of information geometry can be used to equip the parameter space of a theory with a metric.
The central object in this construction is a likelihood function, which is typically intractable once non-perturbative and experimental effects are incorporated.
I will then describe how machine learning can be used to effectively approximate these intractable likelihoods in realistic settings. If time allows I will discuss extensions of these ideas in the context of jet physics and lattice QCD.
Suggested Reading:Better Higgs boson measurements through information geometry arXiv:1612.05261Constraining Effective Field Theories with Machine Learning arXiv:1805.00013 (short PRL)arXiv:1805.00020 (long PRD)QCD-Aware Recursive Neural Networks for Jet Physicshttps://arxiv.org/pdf/1702.00748.pdfJUNIPR: a Framework for Unsupervised Machine Learning in Particle Physicshttps://arxiv.org/pdf/1804.09720.pdfMachine learning action parameters in lattice quantum chromodynamicshttps://arxiv.org/abs/1801.05784
Date & Time
October 24, 2018 | 1:45pm – 2:45pm
Location
Bloomberg Hall Physics LibrarySpeakers
Affiliation
Junior Visiting Professor, School of Natural Sciences; Professor, New York University