Seminar

At all major stages of experimental data processing, especially at the stage of reconstructing the tracks of interacting particles, in the conditions of modern experiments conducted at high-luminosity colliders such as HL-LHC and NICA, major problems have arisen or are expected with achieving results that meet the new requirements for accuracy and speed of obtaining them.

Difficulties in implementing traditional data processing methods are caused primarily by the very high, megahertz frequency of interactions, which leads to an increase in the intensity of the data stream to be processed by orders of magnitude, and in addition, to a significant overlap of event data when they are recorded in track detectors. All these circumstances, especially noticeable at the key stage of processing, which is considered tracking, and designated by physicists as the "Tracking Crisis", have shown that traditionally used tracking algorithms, including the widely used Kalman filter method, demonstrate insufficient efficiency in conditions of high luminosity.

New approaches based on the use of deep neural networks demonstrate their growing potential in solving experimental data processing problems. Due to their high recognition ability, efficient parallelization, and scaling, they are gradually replacing traditional algorithms in high-energy physics applications.

After a brief description of the basics of the theory of artificial neural networks, the report shows the evolution of neural networks from simple perceptrons and Hopfield neural networks to deep neural networks of various types, such as recurrent, graph, transformational, and quantum, taking into account the specifics of modern tracking detectors used in JINR practice and the conditions for recording data from them.

Link to connect
Meeting ID 848 0000 6730
Passcode 734860