The state-of-the-art deep learning (DL) models for jet classification use jet constituent information directly, improving performance tremendously. This draws attention to interpretability, namely, the decision-making process, correlations contributing to the classification, and high-level features (HLFs) representing the difference between signal and background. We address the interpretability issue using a modular architecture called the analysis model (AM), which combines several motivated HLFs as the input. We focus on the generator systematics of the top vs. QCD classification by one of the best classifiers, Particle Transformer (ParT). Taking commonly used event generators Pythia (PY) and Herwig (HW) as examples, we demonstrate that the event weights estimated by the AM generator classifier align the HW classification score distribution to PY ones for QCD jets, with small training uncertainty. This suggests the AM is sufficient to describe simulated QCD jet features with relatively few observables, and generator systematics would also be reduced by reweighting the simulation by data. On the other hand, large event weights are required for QCD-like top jets, which leads to imperfect reweighting for both AM and ParT generator classifiers. Moreover, the AM HLFs are insufficient for describing PY and HW differences, causing lower reweighting accuracy compared with ParT. The missing features are the correlation among the collimated high-energy jet constituents, which are strongly correlated to the energy flow polynomials (EFPs) selected for top vs. QCD classification, showing the complementarity between AM HLFs and the selected EFPs.

Japan

Amon Furuichi

South Korea

United States

Sung Lim

Mihoko Nojiri

Reweighting and analysing event generator systematics by neural networks on high-level features

The system of light electroweakinos and heavy squarks gives rise to one of the most challenging signatures to detect at the LHC. It consists of missing transverse energy recoiled against a few hadronic jets originating either from QCD radiation or squark decays. The analysis generally suffers from the large irreducible $$Z + \textrm{jets}$$ ( $$Z \rightarrow \nu {\bar{\nu }}$$ ) background. In this study, we explore Machine Learning (ML) methods for efficient signal/background discrimination. Our best attempt uses both reconstructed (jets, missing transverse energy, etc.) and low-level (particle-flow) objects. We find that the discrimination performance improves as the $$p_\textrm{T}$$ threshold for soft particles is lowered from 10 to 1 GeV, at the expense of larger systematic uncertainty. In many cases, the ML method provides a factor two enhancement in $$S/\sqrt{S+B}$$ from a simple kinematical selection. The sensitivity on the squark-elecroweakino mass plane is derived with this method, assuming the Run-3 and HL-LHC luminosities. Moreover, we investigate the relations between input features and the network’s classification performance to reveal the physical information used in the background/signal discrimination process.

Poland

France

Rafał Masełek

Kazuki Sakurai

Machine learning electroweakino production

Anomaly detection is a vital technique for exploring signatures of new physics Beyond the Standard Model (BSM) at the Large Hadron Collider (LHC). The vast number of collisions generated by the LHC demands sophisticated deep learning techniques. Similarity learning, a self-supervised machine learning, detects anomalous signals by estimating their similarity to background events. In this paper, we explore the potential of quantum computers for anomaly detection through similarity learning, leveraging the power of quantum computing to enhance the known similarity learning method. In the realm of noisy intermediate-scale quantum (NISQ) devices, we employ a hybrid classical-quantum network to search for heavy scalar resonances in the di-Higgs production channel. In the absence of quantum noise, the hybrid network demonstrates improvement over the known similarity learning method. Moreover, we employ a clustering algorithm to reduce measurement noise from limited shot counts, resulting in 9% improvement in the hybrid network performance. Our analysis highlights the applicability of quantum algorithms for LHC data analysis, where improvements are anticipated with the advent of fault-tolerant quantum computers.

A. Hammad

Masahito Yamazaki

Quantum similarity learning for anomaly detection

Recent advancements in deep learning models have significantly enhanced jet classification performance by analyzing low-level features (LLFs). However, this approach often leads to less interpretable models, emphasizing the need to understand the decision-making process and to identify the high-level features (HLFs) crucial for explaining jet classification. To address this, we consider the top jet tagging problems and introduce an analysis model (AM) that analyzes selected HLFs designed to capture important features of top jets. Our AM mainly consists of the following three modules: a relation network analyzing two-point energy correlations, mathematical morphology and Minkowski functionals for generalizing jet constituent multiplicities, and a recursive neural network analyzing subjet constituent multiplicity to enhance sensitivity to subjet color charges. We demonstrate that our AM achieves performance comparable to the Particle Transformer (ParT) while requiring fewer computational resources in a comparison of top jet tagging using jets simulated at the hadronic calorimeter angular resolution scale. Furthermore, as a more constrained architecture than ParT, the AM exhibits smaller training uncertainties because of the bias-variance tradeoff. We also compare the information content of AM and ParT by decorrelating the features already learned by AM. Lastly, we briefly comment on the results of AM with finer angular resolution inputs.

Jet classification using high-level features from anatomy of top jets

Attention-based transformer models have become increasingly prevalent in collider analysis, offering enhanced performance for tasks such as jet tagging. However, they are computationally intensive and require substantial data for training. In this paper, we introduce a new jet classification network using an MLP mixer, where two subsequent MLP operations serve to transform particle and feature tokens over the jet constituents. The transformed particles are combined with subjet information using multi-head cross-attention so that the network is invariant under the permutation of the jet constituents. We utilize two clustering algorithms to identify subjets: the standard sequential recombination algorithms with fixed radius parameters and a new IRC-safe, density-based algorithm of dynamic radii based on HDBSCAN. The proposed network demonstrates comparable classification performance to state-of-the-art models while boosting computational efficiency drastically. Finally, we evaluate the network performance using various interpretable methods, including centred kernel alignment and attention maps, to highlight network efficacy in collider analysis tasks.

Streamlined jet tagging network assisted by jet prong structure

A beam dump experiment can be seamlessly added to the proposed Inter- national Linear Collider (ILC) program because the high energy electron beam should be dumped after the collision point. The ILC beam dump experiment will provide an excellent opportunity to search for new long-lived particles. Since many of them can be produced by a rare decay of standard model particles, we evaluate spectra of the mesons and τ lepton at the decay based on the PHITS and PYTHIA8 simulations. As a motivated physics case, we study the projected sensitivity of heavy neutral leptons at the ILC beam dump experiment. The heavy neutral leptons can also be produced via deep inelastic scattering and Z boson decay at the ILC main detector, which we include in the projection. With the multi-track signal, the reach would be greatly extended in mass and coupling, even compared with the other proposed searches.

Yasuhito Sakaki

Kohsaku Tobioka

China

Daiki Ueda

First evaluation of meson and τ lepton spectra and search for heavy neutral leptons at ILC beam dump

We study the monojet and dijet channels at the LHC as a tool for searching for squarks and gluinos. We consider two separate R-parity conserving supersymmetric scenarios. In the first scenario we postulate a large mass hierarchy between squarks (  <math> <mover> <mi>q</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q} $$ ) and winos (  <math> <mover> <mi>W</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{W} $$ ), and wino-like neutralino is assumed to be the lightest supersymmetric particle (LSP). The associated squark-wino production, pp →   <math> <mover> <mi>q</mi> <mo>~</mo> </mover> <mover> <mi>W</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q}\overset{\sim }{W} $$ , then leads to a monojet-like signature, where the high p  T  jet is originated from the squark decay,   <math> <mover> <mi>q</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q} $$  → q +   <math> <mover> <mi>W</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{W} $$ . We demonstrate that this associated production, as well as the pp →   <math> <mover> <mi>W</mi> <mo>~</mo> </mover> <mover> <mi>W</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{W}\overset{\sim }{W} $$  + jets production, have a significant impact on the exclusion limit in the squark-neutralino mass plane. The second scenario postulates that the lighter of the squark and gluino is only a few GeV heavier than the LSP neutralino. The associated squark-gluino production, pp →   <math> <mover> <mi>q</mi> <mo>~</mo> </mover> <mover> <mi>g</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q}\overset{\sim }{g} $$ , then leads to a distinctive monojet signature, where the high p  T  jet is produced from the decay of the heavier coloured particle into the lighter one (  <math> <mover> <mi>q</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q} $$  → q+  <math> <mover> <mi>g</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{g} $$  for   <math> <msub> <mi>m</mi> <mover> <mi>q</mi> <mo>~</mo> </mover> </msub> </math>  $$ {m}_{\tilde{q}} $$  &gt;   <math> <msub> <mi>m</mi> <mover> <mi>g</mi> <mo>~</mo> </mover> </msub> </math>  $$ {m}_{\tilde{g}} $$  and   <math> <mover> <mi>g</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{g} $$  → q+  <math> <mover> <mi>q</mi> <mo>~</mo> </mover> </math>  $$ \overset{\sim }{q} $$  for   <math> <msub> <mi>m</mi> <mover> <mi>g</mi> <mo>~</mo> </mover> </msub> </math>  $$ {m}_{\tilde{g}} $$  &gt;   <math> <msub> <mi>m</mi> <mover> <mi>q</mi> <mo>~</mo> </mover> </msub> </math>  $$ {m}_{\tilde{q}} $$ ). The lighter coloured particle is effectively regarded as an invisible particle since the decay products are soft due to the approximate mass degeneracy. We recast existing monojet and dijet analyses and find a non-trivial exclusion limit in the squark-gluino mass plane in this scenario.

Iñaki Lara

Norway

Trygve Buanes

Krzysztof Rolbiecki

Monojet signatures from gluino and squark decays

Deep neural networks trained on jet images have been successful in classifying different kinds of jets. In this paper, we identify the crucial physics features that could reproduce the classification performance of the convolutional neural network in the top jet vs. QCD jet classification. We design a neural network that considers two types of sub-structural features: two-point energy correlations, and the IRC unsafe counting variables of a morphological analysis of jet images. The new set of IRC unsafe variables can be described by Minkowski functionals from integral geometry. To integrate these features into a single framework, we reintroduce two-point energy correlations in terms of a graph neural network and provide the other features to the network afterward. The network shows a comparable classification performance to the convolutional neural network. Since both networks are using IRC unsafe features at some level, the results based on simulations are often dependent on the event generator choice. We compare the classification results of Pythia 8 and Herwig 7, and a simple reweighting on the distribution of IRC unsafe features reduces the difference between the results from the two simulations.

India

Amit Chakraborty

Michihisa Takeuchi

Neural network-based top tagger with two-point energy correlations and geometry of soft emissions

Classification of jets with deep learning has gained significant attention in recent times. However, the performance of deep neural networks is often achieved at the cost of interpretability. Here we propose an interpretable network trained on the jet spectrum S 2(R) which is a two-point correlation function of the jet constituents. The spectrum can be derived from a functional Taylor series of an arbitrary jet classifier function of energy flows. An interpretable network can be obtained by truncating the series. The intermediate feature of the network is an infrared and collinear safe C-correlator which allows us to estimate the importance of an S 2(R) deposit at an angular scale R in the classification. The performance of the architecture is comparable to that of a convolutional neural network (CNN) trained on jet images, although the number of inputs and complexity of the architecture is significantly simpler than the CNN classifier. We consider two examples: one is the classification of two-prong jets which differ in color charge of the mother particle, and the other is a comparison between Pythia 8 and Herwig 7 generated jets.

Interpretable deep learning for two-prong jet classification with jet spectra

We explore signatures related to squark decays in the framework of non-minimally flavour-violating supersymmetry. We consider a simplified model where the lightest squark consists of an admixture of charm and top flavour. By recasting the existing LHC searches for top and charm squarks, we show that the limits on squark masses from these analyses are significantly weakened when the top-charm mixing is sizeable. We propose a dedicated search for squarks based on the $$tc+{E_{\mathrm {T}}^{\mathrm {miss}}}$$ <math><mrow><mi>t</mi><mi>c</mi><mo>+</mo><msubsup><mi>E</mi><mrow><mi>T</mi></mrow><mi>miss</mi></msubsup></mrow></math>  final state which enhances the experimental sensitivity for the case of high mixing, and we map its expected reach for the forthcoming runs of the LHC. We emphasize the role of analyses requiring a jet tagged as produced by the fragmentation of a charm quark in understanding the squark mixing pattern, thus providing a novel handle on new physics. Our results show that, in order to achieve full coverage of the parameter space of supersymmetric models, it is necessary to extend current experimental search programmes with analyses specifically targeting the cases where the lightest top-partner is a mixed state.