Jet-images — deep learning edition

By Luke de Oliveira, Michael Kagan, Lester W. Mackey, Benjamin Nachman, Ariel Schwartzman. 2015

344 citations recorded by the source corpus.

Building 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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Nuclear and High Energy Physics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Nuclear and High Energy Physics

Citations in the corpus

Learning Particle Physics by Example: Location-Aware Generative Adversarial Networks for Physics Synthesis (2017, methodology): “[20] except now there is not only signal and background classes, but also fake and real categories.”
Energy flow polynomials: a complete linear basis for jet substructure (2017, methodology): “methods. On the other hand, multivariate methods can sometimes obscure the specic physics information that the model learns, leading to recent eorts to \open the box" of machine learning tools [63,72,113,132,133]. Even with an open box, though, theoretical calculations of multivariate distribut”
Efficient Probabilistic Inference in the Quest for Physics Beyond the Standard Model (2018, background): “The field is increasingly employing deep learning techniques allowing these algorithms to process high-dimensional, low-level data [14, 17, 31, 55, 75].”
JUNIPR: a framework for unsupervised machine learning in particle physics (2018, background): “ork methods. { 2 { adopted instead, and they generally outperform fully-connected networks due to their ecient feature detection. In the rst applications of CNNs to jet images, on boosted Wdetection [20] and quark/gluon discrimination [21], it was indeed found that simple CNNs could generally outper”
The Lund jet plane (2018, methodology): “ing a batch size of 128, with Adam as optimisation algorithm [78] and a categorical cross-entropy loss function. The parameters for the machine learning are similar, where relevant, to those in Refs. [37,38], though we used a greater number of pixels for the images and larger networks. Training is c”
Deep learning in color: towards automated quark/gluon jet discrimination (2016, methodology): “33 pixels corresponds to a = ˚= 0:024. The discretization into ˚= 0:024 0:024 pixels also acts as a kind of coarse-graining often used as a primitive detector simuilator (e.g. see [44]). We follow [12] in using transverse momentum as our pixel intensity rather than the energy due to the invariance ”
Energy flow networks: deep sets for particle jets (2018, methodology): “parable classication performance with similar information. The other machine learning architectures we compare to are a DNN trained on the Nsubjettiness basis [23,67,68], a CNN trained on jet images [4,5,8], and a linear classier trained on the energy ow basis [28]. Their performance, as given by th”
Jet Substructure Classification in High-Energy Physics with Deep Neural Networks (2016, methodology): “[28] Luke de Oliveira, Michael Kagan, Lester Mackey, Benjamin Nachman, and Ariel Schwartzman.”
A Transdisciplinary Review of Deep Learning Research and Its Relevance for Water Resources Scientists (2017, background): “Back to HEP, de Oliveira et al. (2016) visualized the convolutional layer activations to understand what unique information from jet images was learned by the network to distinguish different particles.”
ParticleNet: Jet Tagging via Particle Clouds (2019, background): “, W boson tagging [25–29, 35], top tagging [32–34] and quark-gluon tagging [30, 31].”
Searching for new physics with deep autoencoders (2018, background): “iness? We saw here how the CNN autoencoder was reasonably decorrelated with mass. It would be interesting to explore ways to more explicitly decorrelate in mass. The \variable planing" ideas of [3,47] may be useful in this context. Or one could envision training an ensemble of autoencoders on j”
QCD-aware recursive neural networks for jet physics (2017, background): “In agreement with previous results based on deep learning [2, 6], we see that our RNN classifier clearly outperforms this variable.”
Machine learning and the physical sciences (2019, methodology): “Shortly after the first applications of deep learning for event selection, deep convolutional networks were used for the purpose of jet tagging, where the low-level detector data lends itself to an image-like representation (Baldi et al., 2016a; de Oliveira et al., 2016).”