Artificial intelligence learns to predict elementary particle signals

Experiments in high energy physics require work with Big Data. For example, at the LHC, millions of collisions occur every second and detectors record these particles and determine their characteristics. But to receive an accurate analysis of the experimental data, it is necessary to know how the detector reacts to known particles. To do this, you usually use special software configured for the geometry and physics of a particular detector.

Such packages provide a fairly accurate description of the response of the carrier to the passage of charged particles, but the generation rate of each event can be very slow. In particular, the simulation of a single LHC event can last several seconds. Since millions of charged particles collide every second in the collider itself, an exact description becomes inaccessible.

Researchers at HSE and Yandex Data Analysis School have been able to speed up the simulation with Generative Adversarial Networks. These are composed of two neural networks that compete against each other during competitive training. This training method is used, for example, to generate photos of people who do not exist. A network learns to create images similar to reality and the other seeks to find differences between artificial and real representations.

"It's amazing how the methods developed to generate realistic cat pictures allow us to accelerate physical computations by several orders of magnitude," notes Nikita Kaseev, Ph.D. student at HSE and co – author of the study.

Researchers have formed competitive generative networks to predict the behavior of charged elementary particles. The results showed that physical phenomena can be described using neural networks with very high accuracy.

"The use of generative competitive networks to quickly simulate detector behavior will certainly facilitate future experiments," says Denis Derkach, an assistant professor at the Faculty of Computer Science and co-author of the study. "We have essentially used the most modern training methods available in data science and our knowledge of detector physics, and the diversity of our team of data scientists and physicists has also made this possible."

Explore further:
International team of physicists continues research on new areas of physics

More information:
Denis Derkach et al, Cherenkov Detects Rapid Simulation Using Neural Networks, Nuclear Instruments and Research Methods in Physics Section A: Accelerators, Spectrometers, Detectors and Associated Equipment (2019). DOI: 10.1016 / j.nima.2019.01.031