MLIT at the School dedicated to the application of novel data processing and analysis methods On 4 December, the second Scientific School “New Methods for Processing Data of a Physical Experiment” ended at Moscow Institute of Physics and Technology (MIPT). The event aimed to combine the efforts of various scientific groups in elaborating novel concepts for processing large data volumes using artificial intelligence (AI) methods, including machine learning methods, artificial neural networks, soft computing, etc. The School was organized by MIPT and the Joint Institute for Nuclear Research (JINR). A number of specialists from the Meshcheryakov Laboratory of Information Technologies (MLIT) of JINR participated in the event and delivered talks. At the School’s opening, MLIT Director Sergei Shmatov welcomed the participants on behalf of the Joint Institute. He explained why physicists need to master computational methods and data processing techniques using the example of experiments at the Large Hadron Collider (LHC) at CERN. The talk covered MLIT plans for organizing systems to collect, process and store data within the NICA (Nuclotron based Ion Collider Acility) megascience project. A report by MLIT Senior Researcher Martin Bures on the application of quantum algorithms to event reconstruction in collider experiments was also presented in the School’s program. The School is aimed at specialists, postgraduates and students of leading Russian institutes interested in the application of new methods for acquiring, processing and handling data from physical experiments of the megascience level. One of its objectives is to provide an overview of novel methods for working with data from a modern physical experiment, to detect problems that require solutions and elaborate approaches to solving them. “Artificial intelligence methods are becoming more and more common in a wide variety of areas of our lives. However, for a number of fields of physics, they are still largely new. Within this School, we want to acquaint young people with AI methods that are currently employed in physics, to interest them and give them an overview, to show where there is a place for new methods. At the same time, our task is to orient young scientists and, without giving them inflated expectations from AI, show entry points to specific tasks in this area and practical application opportunities”, Sergei Shmatov noted. “The exponential growth of data volumes in fundamental physics experiments entails novel approaches to analysis. Machine learning and other advanced technologies are crucial in order to overcome these challenges and achieve breakthrough results,” VBLHEP Senior Researcher Alexey Aparin, one of the School’s organizers, commented on the objectives of the event. The experience of various scientific groups in implementing AI methods at different stages of experimental data processing, as well as the monitoring of the parameters of detectors and the accelerator complex during operation, was considered within the School. All this can be used in the future in the functioning of megascience experimental facilities, such as the NICA complex at JINR. This is already the second School, and its organizers aim to make the event traditional with the further involvement of a wider circle of Russian and international research and scientific-educational organizations and scientific groups. “We see the development of the School not only in expanding the geography of lecturers, but also in the participants’ deeper immersion in the practice and methodology of applying AI. By supplementing the program of overview talks with practical classes, we will be able to demonstrate to the participants how exactly the methods presented in the talks are employed and what results this gives. Most importantly, the participants will have the opportunity to try it all by themselves in practice. Ideally, based on the experience of holding the MLIT IT School, we want to enlarge the program with hackathons and tutorials,” MLIT Deputy Director Nikolay Voytishin indicated. The School’s program embraced reports on key aspects of the development of computing technologies in high-energy physics, as well as on general approaches to the application of generative neural networks to solve data processing tasks in high-energy physics and algorithms for searching for B-meson decays based on convolutional neural networks. Big Data analysis methods in particle physics and astrophysics, in particular, for the Baikal-GVD neutrino experiment, current issues related to the organization of work with data and practical solutions for accelerating analysis and reducing the time spent on debugging working programs were considered. A number of talks were devoted to the application of machine learning methods to some tasks of data selection and classification within the BM@N and SPD experiments at NICA, as well as to the calculation of magnetic fields in accelerators and the maintenance of the required parameters of the accelerated particle beam. “Daily final sessions dedicated to general discussions and the discussion of the reports presented during the day became a valuable part of the School,” Alexey Aparin said. “During that time, the participants were able to continue communicating in a less formal manner on issues of interest that were discussed in the lectures. It is noteworthy that many of the speakers are young scientists themselves, which creates the most friendly atmosphere for the discussion. In this way, we hope to enhance students’ interest in working in megascience experiments and attract even more young people to physics.” Over 80 specialists, postgraduates and students from leading Russian research and educational organizations, including JINR, MIPT, NRC KI, HSE University, INR RAS and SPbSU, participated in the three-day School “New Methods for Processing Data of a Physical Experiment”. Artificial intelligence methods at MLIT Speaking about the range of activities at MLIT carried out using AI methods, Sergei Shmatov underlined that the ideologist of this direction in the Laboratory was Gennady Ososkov, Chief Researcher of the MLIT Scientific Department of Computational Physics. Under his supervision, a group of scientists works in the field of applying machine learning and neural network methods for tracking, namely, charged particle trajectory reconstruction, as well as in the direction of quantum computing based on quantum annealing. At MLIT, AI methods are actively employed in particle physics, where the NICA project and its experiments (MPD, SPD, BM@N) are essential. MLIT also steadily applies AI methods for the CMS experiment at the LHC (CERN), using neural networks for experimental data modeling and analysis. Work on elementary particle identification, as well as on the application of neural networks in numerical integration, is performed in the Sector of Nonlinear Systems Modeling Methods headed by Alexander Ayriyan. The MLIT Sector of Heterogeneous Computing and Quantum Informatics under the guidance of Oksana Streltsova employs AI methods in the field of radiobiology within interlaboratory research jointly with the JINR Laboratory of Radiation Biology (LRB) and Serbian scientific organizations. This work is also carried out in the direction of algorithm development based on machine and deep learning, as well as of web service elaboration for automating the processing of data obtained in experiments conducted at LRB to study the effects of ionizing radiation on biological objects, namely, services for analyzing data from behavioral tests on small laboratory animals have been developed, and a web service for the automated analysis of radiation-induced foci in cell nuclei has been created. Impressive results have been obtained by Alexander Uzhinskiy, Senior Researcher at the Scientific and Technical Department of External Communications and Distributed Information Systems. Using convolutional neural networks in image recognition, the team of this project has created an online platform and a mobile application capable of detecting various plant diseases and pests with very high accuracy. The active implementation of AI methods (neural network approach, machine and deep learning methods and algorithms, etc.) to solve a wide range of tasks is driven by many factors. The key ones involve the development of computing architectures, especially when applying deep learning methods for convolutional neural networks, as well as the development of libraries that implement a wide variety of algorithms and frameworks that enable to quickly build different neural network models. To provide all these capabilities both for developing mathematical models and algorithms, and for performing resource-intensive computing, including on graphics accelerators, which allow one to significantly reduce the calculation time, an ecosystem for machine and deep learning tasks, as well as for data analysis tasks, has been created for users of the HybriLIT platform and is actively developing. A quantum polygon enabling to develop algorithms using quantum simulators has also been deployed on the resources of the ecosystem. The HybriLIT heterogeneous computing platform is part of the Multifunctional Information and Computing Complex (MICC) of MLIT JINR. The platform is a multicomponent system, with the major computing resource being the “Govorun” supercomputer with a peak performance of 1.7 PFlops for double-precision calculations and 26 PFlops with half-precision for artificial intelligence tasks. The specifics of working with data in physics experiments were discussed at MIPT