Benjamin Philip Nachman

Jet-images — deep learning edition

A bstractBuilding on the notion of a particle physics detector as a camera and the collimated streams of high energy particles, or jets, it measures as an image, we investigate the potential of machine learning techniques based on deep learning architectures to identify highly boosted W bosons. Modern deep learning algorithms trained on jet images can out-perform standard physically-motivated feature driven approaches to jet tagging. We develop techniques for visualizing how these features are learned by the network and what additional information is used to improve performance. This interplay between physicallymotivated feature driven tools and supervised learning algorithms is general and can be used to significantly increase the sensitivity to discover new particles and new forces, and gain a deeper understanding of the physics within jets.

Learning Particle Physics by Example: Location-Aware Generative Adversarial Networks for Physics Synthesis

We provide a bridge between generative modeling in the Machine Learning community and simulated physical processes in high energy particle physics by applying a novel Generative Adversarial Network (GAN) architecture to the production of jet images—2D representations of energy depositions from particles interacting with a calorimeter. We propose a simple architecture, the Location-Aware Generative Adversarial Network, that learns to produce realistic radiation patterns from simulated high energy particle collisions. The pixel intensities of GAN-generated images faithfully span over many orders of magnitude and exhibit the desired low-dimensional physical properties (i.e., jet mass, n-subjettiness, etc.). We shed light on limitations, and provide a novel empirical validation of image quality and validity of GAN-produced simulations of the natural world. This work provides a base for further explorations of GANs for use in faster simulation in high energy particle physics.

Bisection-based asymmetric MT2 computation: a higher precision calculator than existing symmetric methods

A bstractAn MT2 calculation algorithm is described. It is shown to achieve better precision than the fastest and most popular existing bisection-based methods. Most importantly, it is also the first algorithm to be able to reliably calculate asymmetric MT2 to machine-precision, at speeds comparable to the fastest commonly used symmetric calculators.

Weakly supervised classification in high energy physics

A bstractAs machine learning algorithms become increasingly sophisticated to exploit subtle features of the data, they often become more dependent on simulations. This paper presents a new approach called weakly supervised classification in which class proportions are the only input into the machine learning algorithm. Using one of the most challenging binary classification tasks in high energy physics — quark versus gluon tagging — we show that weakly supervised classification can match the performance of fully supervised algorithms. Furthermore, by design, the new algorithm is insensitive to any mis-modeling of discriminating features in the data by the simulation. Weakly supervised classification is a general procedure that can be applied to a wide variety of learning problems to boost performance and robustness when detailed simulations are not reliable or not available.

Jets from jets: re-clustering as a tool for large radius jet reconstruction and grooming at the LHC

A bstractJets with a large radius R ≳ 1 and grooming algorithms are widely used to fully capture the decay products of boosted heavy particles at the Large Hadron Collider (LHC). Unlike most discriminating variables used in such studies, the jet radius is usually not optimized for specific physics scenarios. This is because every jet configuration must be calibrated, insitu, to account for detector response and other experimental effects. One solution to enhance the availability of large-R jet configurations used by the LHC experiments is jet re-clustering. Jet re-clustering introduces an intermediate scale r < R at which jets are calibrated and used as the inputs to reconstruct large radius jets. In this paper we systematically study and propose new jet re-clustering configurations and show that re-clustered large radius jets have essentially the same jet mass performance as large radius groomed jets. Jet re-clustering has the benefit that no additional large-R calibration is necessary, allowing the re-clustered large radius parameter to be optimized in the context of specific precision measurements or searches for new physics.

Classification without labels: Learning from mixed samples in high energy physics

A bstractModern machine learning techniques can be used to construct powerful models for difficult collider physics problems. In many applications, however, these models are trained on imperfect simulations due to a lack of truth-level information in the data, which risks the model learning artifacts of the simulation. In this paper, we introduce the paradigm of classification without labels (CWoLa) in which a classifier is trained to distinguish statistical mixtures of classes, which are common in collider physics. Crucially, neither individual labels nor class proportions are required, yet we prove that the optimal classifier in the CWoLa paradigm is also the optimal classifier in the traditional fully-supervised case where all label information is available. After demonstrating the power of this method in an analytical toy example, we consider a realistic benchmark for collider physics: distinguishing quark- versus gluon-initiated jets using mixed quark/gluon training samples. More generally, CWoLa can be applied to any classification problem where labels or class proportions are unknown or simulations are unreliable, but statistical mixtures of the classes are available.

Sneaky light stop

A light supersymmetric top quark partner (stop) with a mass nearly degenerate with that of the standard model (SM) top quark can evade direct searches. The precise measurement of SM top properties such as the cross-section has been suggested to give a handle for this ‘stealth stop’ scenario. We present an estimate of the potential impact a light stop may have on top quark mass measurements. The results indicate that certain light stop models may induce a bias of up to a few GeV, and that this effect can hide the shift in, and hence sensitivity from, cross-section measurements. Due to the different initial states, the size of the bias is slightly different between the LHC and the Tevatron. The studies make some simplifying assumptions for the top quark measurement technique, and are based on truth-level samples.

Pileup Mitigation with Machine Learning (PUMML)

A bstractPileup involves the contamination of the energy distribution arising from the primary collision of interest (leading vertex) by radiation from soft collisions (pileup). We develop a new technique for removing this contamination using machine learning and convolutional neural networks. The network takes as input the energy distribution of charged leading vertex particles, charged pileup particles, and all neutral particles and outputs the energy distribution of particles coming from leading vertex alone. The PUMML algorithm performs remarkably well at eliminating pileup distortion on a wide range of simple and complex jet observables. We test the robustness of the algorithm in a number of ways and discuss how the network can be trained directly on data.

Convolved substructure: analytically decorrelating jet substructure observables

A bstractA number of recent applications of jet substructure, in particular searches for light new particles, require substructure observables that are decorrelated with the jet mass. In this paper we introduce the Convolved SubStructure (CSS) approach, which uses a theoretical understanding of the observable to decorrelate the complete shape of its distribution. This decorrelation is performed by convolution with a shape function whose parameters and mass dependence are derived analytically. We consider in detail the case of the D2 observable and perform an illustrative case study using a search for a light hadronically decaying Z′. We find that the CSS approach completely decorrelates the D2 observable over a wide range of masses. Our approach highlights the importance of improving the theoretical understanding of jet substructure observables to exploit increasingly subtle features for performance.

Investigating multiple solutions in the constrained minimal supersymmetric standard model

A bstractRecent work has shown that the Constrained Minimal Supersymmetric Standard Model (CMSSM) can possess several distinct solutions for certain values of its parameters. The extra solutions were not previously found by public supersymmetric spectrum generators because fixed point iteration (the algorithm used by the generators) is unstable in the neighbourhood of these solutions. The existence of the additional solutions calls into question the robustness of exclusion limits derived from collider experiments and cosmological observations upon the CMSSM, because limits were only placed on one of the solutions. Here, we map the CMSSM by exploring its multi-dimensional parameter space using the shooting method, which is not subject to the stability issues which can plague fixed point iteration. We are able to find multiple solutions where in all previous literature only one was found. The multiple solutions are of two distinct classes. One class, close to the border of bad electroweak symmetry breaking, is disfavoured by LEP2 searches for neutralinos and charginos. The other class has sparticles that are heavy enough to evade the LEP2 bounds. Chargino masses may differ by up to around 10% between the different solutions, whereas other sparticle masses differ at the sub-percent level. The prediction for the dark matter relic density can vary by a hundred percent or more between the different solutions, so analyses employing the dark matter constraint are incomplete without their inclusion.