MLIT young researchers’ results presented to PAC PP participants At the 62nd meeting of the JINR Program Advisory Committee for Particle Physics which took place on 23 June 2025, six young scientists from the Meshcheryakov Laboratory of Information Technologies presented the results of their work in the format of poster reports. GRAPH NEURAL NETWORK WITH ATTENTION AND TWO-STAGE AGGREGATION FOR PARTICLE TRACK RECONSTRUCTION IN THE MPD TPC OF THE NICA COMPLEX Yauheni Talochka, Gennady Ososkov, Nikolay Voytishin One of the main challenges for the TPC MPD at the NICA accelerator complex, under the extremely high interaction frequency reaching megahertz levels, which significantly increases the volume of data recorded by the detectors, is particle track reconstruction. We present a Graph Neural Network (GNN) with an attention mechanism and two-stage aggregation for particle track reconstruction in the NICA MPD TPC. A dataset of 1,000 Au-Au collision events generated with MPDRoot was utilized to train the GNN model. We demonstrate that the introduction of the attention mechanism and two-stage aggregation allows improving the performance of the GNN in edge classification, achieving 96.2% accuracy and 92.6% in both purity and efficiency. Meanwhile, the track reconstruction efficiency exceeds 90% for track integrities below 80%, but drops sharply to 48% for higher integrity thresholds. SEPARATION OF PARTICLE TRAJECTORIES BY BEAM COLLISION EVENTS ACCUMULATED IN A SINGLE TIME SLICE IN THE SPD DETECTOR AT THE NICA ACCELERATOR USING GRAPH NEURAL NETWORKS Saveliy Omelyanchuk, Yauheni Talochka, Gennady Ososkov The work is devoted to the development of deep learning methods for the classification of elementary particle tracks. The architecture of a graph neural network (GNN) for track classification by events in each time slice in the SPD experiment is explored. A new approach to track sorting, along with an investigation of learning dynamics and model testing under various conditions, is presented. The model is implemented and trained using modern deep learning tools that enable parallel tensor computations. SPD OFFLINE DATA PROCESSING SYSTEM FOR THE SPD EXPERIMENT Artem Petrosyan, Alexey Konak, Nikita Monakov The SPD (Spin Physics Detector) facility at the NICA accelerator complex at JINR is under construction. In addition to the facility itself, the software for the future experiment is also being developed. There is already a constant demand for sufficiently large-scale data productions to simulate physical processes in the future experiment. To facilitate their implementation, MLIT specialists are developing a set of systems and services that allow for the orderly storage and processing of experimental data both on JINR resources and on the resources of the institutes that are members of the SPD collaboration. The presented systems and services are in trial operation. Over the past six months, the system has modeled more than 1 billion physical events for the benefit of the experiment’s physical groups and obtained over 200 TB of data. SPD ONLINE FILTER MIDDLEWARE Nikita Greben, Leonid Romanychev, Danila Oleynik, Artem Plotnikov, Polina Korshunova SPD Online Filter is a hardware-software system designed for the multi-stage, high-throughput processing of data from the SPD detector. Its main task is primary data processing to reduce the data volume for long-term storage and subsequent full processing. SPD Online Filter comprises a dedicated computing cluster, middleware, and a set of application-level services. The middleware layer consists of three microservice-based systems that communicate via lightweight API gateways for request routing and a message broker to decouple producers and consumers. Together, they form a configurable, fault-tolerant, and scalable data-processing pipeline. The poster illustrates the architecture of the overall system and its constituent subsystems, demonstrates coordinated interaction between the components, and shows how they work together to deliver the reliable, scalable, real-time processing of raw data to meet the SPD experiment’s requirements. ANALYSIS OF THE DRELL–YAN PROCESS BASED ON RUN2 OPEN DATA FROM THE CMS EXPERIMENT Yurii Korsakov, Sergei Shmatov, Alexander Lanyov (VBLHEP) The results of the analysis of the Drell-Yan process using CMS open data for 2015 are presented. The dataset corresponds to an integrated luminosity of 2.64 fb-1 at a center-of-mass energy of 13 TeV. The differential cross section of the Drell-Yan process and the kinematic properties of the muon pair were calculated. SEARCH FOR DARK MATTER PARTICLES PREDICTED BY THE SCALAR PORTAL MODEL WITH A TWO-DOUBLET EXPANSION OF THE HIGGS SECTOR AND ONE EXTRA SINGLET Maria Savina (BLTP), Kirill Slizhevskii, Sergei Shmatov The CMS experiment at the LHC investigates the possibility of physics beyond the Standard Model. To search for candidates for dark matter (DM) particles, Z boson events with a large missing transverse momentum in the final state are used. The analysis is based on data at a center-of-mass energy of 13 TeV collected during CMS RUN2, which corresponds to an integrated luminosity of 137 fb-1. The results of the analysis enabled to constrain the parameters of simplified DM models with vector, axial-vector, scalar, pseudoscalar mediators and a two-doublet Higgs model (2HDM+a) with an additional pseudoscalar mediator. We present a simulation of signal processes for 2HDM+S/a performed using the MadGraph5 and Pythia8 generators, which are used in the analysis of CMS RUN3 data with an energy of 13.6 TeV.
