Towards machine learning analytics for jet substructure

Abstract

The past few years have seen a rapid development of machine-learning algorithms. While surely augmenting performance, these complex tools are often treated as black-boxes and may impair our understanding of the physical processes under study. The aim of this paper is to move a first step into the direction of applying expert-knowledge in particle physics to calculate the optimal decision function and test whether it is achieved by standard training, thus making the aforementioned black-box more transparent. In particular, we consider the binary classification problem of discriminating quark-initiated jets from gluon-initiated ones. We construct a new version of the widely used N-subjettiness, which features a simpler theoretical behaviour than the original one, while maintaining, if not exceeding, the discrimination power. We input these new observables to the simplest possible neural network, i.e. the one made by a single neuron, or perceptron, and we analytically study the network behaviour at leading logarithmic accuracy. We are able to determine under which circumstances the perceptron achieves optimal performance. We also compare our analytic findings to an actual implementation of a perceptron and to a more realistic neural network and find very good agreement.

Jet tagging in the Lund plane with graph networks

“…This topic has been the focus of much attention over the past decade, with a range of approaches being developed to extract information from a jet's substructure [2][3][4][5][6]. In recent years, a new generation of tools based on deep learning models have emerged, which can achieve very high performance on specific benchmarks [7][8][9][10][11][12][13][14][15][16][17][18][19][20] and provide some insights into what kinematic variables drive the discrimination performance [21][22][23][24][25][26][27][28][29][30]. A limitation of such deep learning-based methods is the difficulty to estimate their uncertainties, as well as their proneness to rely on unphysical features present in the training data to achieve their high performance, as this data is generally derived from Monte Carlo simulations of proton collisions.…”

Quantum-inspired event reconstruction with Tensor Networks: Matrix Product States

“…Due to different particle clusters occurring on the η − φ-plane, each pixel is only closely correlated with the close-by pixels. The η − φ-plane can be 6 Frobenius norm is defined as…”

“…Due to the requirement of analysing vast, highly correlated data in order to exploit the full physics potential of the LHC, it becomes more and more important to develop a fundamental understanding of the data analysis methods applied. Jet substructure analysis is a particularly popular research area where analytic reconstruction techniques [1][2][3][4][5][6] co-exist with numerical multivariate analyses methods [7][8][9][10][11]. The combination of a large amount of available data with an excellent theoretical understanding of the underlying physics in collider phenomenology provides the ideal environment to explore novel reconstruction techniques and to improve our understanding of existing approaches.…”

Topological obstructions to autoencoding

“…This manifold, which describes the energies and momenta of particles produced in relativistic collisions, has dimension 3n − 4 for n final-state particles, and has a natural embedding in R 4n whose coordinates comprise the n final-state 4-vectors. Training data for a machine learning task derived from these 4vectors, whether low-level [1,2,[8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] or high-level [24][25][26][27][28][29][30], must still at some level inherit the geometry and topology of phase space [31][32][33][34][35][36][37][38][39][40]. In this context, we can make the notion of "anomaly" more precise.…”