Our laboratory "HPC&BigData" is mainly engaged in two areas:

1. research and engineering of effective numerical codes for simulating various physical problems of fundamental and applied nature, as well as their subsequent analysis,
2. analysis of current solutions in the area of hardware accelerators, as well as the maintenance and monitoring of cluster supercomputers.

What is more, computational efficiency can be considered as both the acceleration of numerical calculations within a single laptop and the use of massive parallelism and graphics cards for computing on a cluster.

In addition, we are studying the applicability of new approaches (optimization, neural networks) to solving a number of problems in applied physics. In the field of computational physics, the main emphasis is placed on plasma physics and laser physics. Namely, scientists investigate the issues of interaction of laser radiation with matter or electron beams are investigated to generate attosecond pulses or create new sources of X-ray and gamma radiation. In particular, recently we have been actively looking for new methods for generating swirling (having angular momentum) laser pulses.

The possibility of generating synchrotron radiation and attosecond pulses from more compact sources is considered. In addition, some students are engaged in numerical calculations in the field of fluid dynamics. Among the engineering tasks, it is worth highlighting the implementation of neural networks on programmable controllers of the FPGA type, as well as the optimization of the task scheduler of the Zhores supercomputer to perform the specified calculations with the most efficient utilization of the computing power of the cluster.

Since the field of nuclear photonics is actively developing and there are still many interesting problems in it, we plan to continue research in this area using classical numerical methods and try neural network approaches in special cases.

We should also note the development of our own optimized code for modeling the interaction of laser radiation with plasma. Such an approach, if successfully developed, will reduce the cost of numerical modeling of such problems by tens of thousands of times.

AI has long been actively used in the processing of experimental results (for example, the search for new particles at the Large Hadron Collider). And since not so many experiments have been carried out in the field of nuclear photonics, there is not enough real data for a similar use of AI. It seems very promising, instead of real experiments, to carry out a large number of numerical calculations and train models already on simulation data.