2023 Future Collider Seminar Series



Seminars Home > Seminars


4b+X/4B+W via electroweak multi-Higgs production as smoking gun signals for the Type-I 2HDM at the LHC


2023 May 2, 17:00-18:30, 1423호

Speaker: Sanyal, Prasenjit (Konkuk Univ.)

Online: https://indico.kias.re.kr/event/205/


Abstract : The existence of additional Higgs bosons, besides the one discovered by the LHC, already a decade ago, is predicted by most frameworks of new physics. Observation of a second Higgs boson (charged or neutral) will thus provide firm evidence that the underlying manifestation of the Electroweak Symmetry Breaking (EWSB) mechanism is a non-minimal one. The majority of analyses, both phenomenological and experimental ones, involving additional Higgs bosons concentrate on QCD induced production modes, However, the QCD induced processes are not necessarily to be high in new physics models owing to the non-standard couplings of the new Higgs bosons to the fermions and gauge bosons. As a reference we take the Type-I two Higgs doublet model (2HDM) as a simple extension of Standard Model where the Electroweak (EW) processes dominate over the QCD processes. 

I would like to discuss a full detector-level Monte Carlo analysis to establish that the inclusive 4b + X final state via EW processes can provide simultaneous reconstruction of all the additional Higgs boson masses. I will present the algorithms for the mass reconstructions of the additional Higgs bosons. However, the challenge is to tame the large multi-jet background. As an alternative, I will discuss the 4b + W channel which can be mediated only through the EW process. The leptonic decay of the W boson and therefore the requirement of at least one lepton at the detector level is useful to evade the multi-jet background. The χ 2 variable constructed out of the signal topology can be a very useful selection criteria to reduce the t ¯t background.



A Review of Machine Learning Applications on Jet Tagging


2023 June 8, 16:00-17:30, 8101호

Speaker: Wang, Daohan (Konkuk Univ.)

Online: https://indico.kias.re.kr/event/198/


Abstract : Jet tagging is an important task in high-energy physics, where the goal is to identify jets of particles produced in particle collisions. Machine learning has proven to be a powerful tool in this field, allowing for improved accuracy and efficiency in jet tagging. In this talk, we provide a comprehensive review of the state-of-the-art machine learning techniques used in jet tagging. We categorize these techniques into three main representation types: image-based, particle-based, and point cloud-based. For each representation, we discuss the corresponding neural network architectures, including CNNs, 1D-CNNs, RNNs, Deep set frameworks, ParticleNets, ABCNets, LorentzNets, and Transformers. We also introduce a new architecture which incorporates the pairwise particle interaction and the pairwise jet feature interaction to the Point Cloud Transformer , called P-DaViT. This talk aims to provide an overview of the current landscape of machine learning applications in jet tagging.



Langevin simulation of dark matter kinetic eqilibration


2023 June 13, 17:00-18:30, 1423호

Speaker: Mikko, Laine (AEC, Bern Univ.)

Online: https://indico.kias.re.kr/event/199/


Abstract : Dark matter computations based on the freeze-out mechanism regularly assume the presence of kinetic equilibrium, however recently the validity of this assumption has been questioned in specific cases. We show how kinetic equilibration can be studied numerically, with the help of Langevin simulations. A benchmark point is taken from the scalar singlet model in the resonant regime (scalar mass 60 GeV). For fixed couplings, effects on the 40% level can be found for the final abundance.



Multi-photon decays of the Higgs boson at the LHC


2023 June 28, 16:00-17:30, 1424호 

Speaker: Samuel D. Lane (KAIST)

Online: https://indico.kias.re.kr/event/212/


Abstract : Many new physics scenarios predict multi-photon Higgs resonances. One such scenario is the dark axion portal model. The primary decay chain that we study is  h →γD γD → 2 γ2 a → 6γ. However, depending on the relevant kinematics, the photons can become well-collimated and appear as photon-jets (multiple photons that appear as a single photon in the detector) or ξ-jets (non-isolated multi-photon signals that do not pass the isolation criterion). These effects cause the true six-photon resonance to appear as other multi-photon signals, such as two and four photons. We classify the mass regions where two, four, and six-photon resonances dominate. The four-photon signal is particularly interesting. These events mainly occur when the photons from the axion-like particles are collimated into photon-jets. The decay of the dark photon is then γD → γa → γ + γ-jet, which is an apparent violation of the Landau-Yang theorem. We show that current measurements of h → 2γ and searches for h → 4γ at the Large Hadron Collider (LHC) can limit BR(h → γD γD ) ≲ 10−3. This model also motivates new searches for Higgs decays into six isolated photons or ξ-jets at the LHC. While there are currently no dedicated searches, we show that many of the Higgs to six isolated photons or ξ-jet events could pass two or three-photon triggers. That is, new physics could be found by reanalyzing existing data. These multi-photon signals provide excellent footing to explore new physics at the LHC and beyond.



MadNIS - Neural networks for multi-channel integration


2023 June 30, 16:00-17:30, 1423호

Speaker: Winterhalder, Ramon (CP3, Louvain Univ.)

Online: https://indico.kias.re.kr/event/207/


Abstract : Theory predictions for the LHC require precise numerical phase-space integration and generation of unweighted events. We combine machine-learned multi-channel weights with a normalizing flow for importance sampling to improve classical methods for numerical integration.  We develop an efficient bi-directional setup based on an invertible network, combining online and buffered training for potentially expensive integrands. We illustrate our method for the Drell-Yan process with an additional narrow resonance. In addition to these results from our paper (https://arxiv.org/abs/2212.06172), MadNIS now interfaces with MadGraph, and I will present preliminary results from our upcoming comparison between MadNIS and classical MadGraph for various LHC processes.



Structures of Neural Network Effective Theories


2023 August 1, 10:00-11:30, 1423호

Speaker: Zhang, Zhengkang (USU, Utah State, USA)

Online: https://indico.kias.re.kr/event/226/


Abstract : I will discuss an intriguing connection between neural networks and effective field theories, and introduce a diagrammatic approach to perform calculations in these theories which makes transparent the structures of RG flows underlying the success of critical tuning of deep neural networks. Based on arXiv:2305.02334.



Lattice U(N) gauge theory in the strong coupling limit on D-wave quantum computer


2023 August 29, 16:00-17:30, 8101호

Speaker: Kim, Jangho (Forschungszentrum Julich)

Online: https://indico.kias.re.kr/event/242/


Abstract : Quadratic Unconstrained Binary Optimization (QUBO) problems can be addressed on quantum annealing systems. We reformulate the strong coupling lattice QCD dual representation as a QUBO matrix. We confirm that importance sampling is feasible on the D-Wave Advantage quantum annealer. We describe the setup of the system and present the first results obtained on a D-wave quantum computer for U(N) gauge group.