Scientists Use Generative Adversarial Networks to Accelerate Simulation of Processes at LHC

Researchers at HSE University and Yandex have devised a technique that speeds up the simulation of processes at the Large Hadron Collider (LHC). The outcomes of the study were reported in Nuclear Instruments and Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment.

High-energy physics experiments involve working with big data. For instance, at the LHC, each second, millions of collisions take place, and these particles are registered and their characteristics determined by detectors. However, to achieve an accurate analysis of experimental data, it is essential to understand the way the detector reacts to known particles. In general, this is performed using special software configured for the geometry and physics of a specific detector.

Packages such as these offer a reasonably precise description of the response of the medium to the passage of charged particles; however, it is probable for the rate of generation of each event to be very gradual. Specifically, the simulation of the single LHC event could require several seconds. In the collider, since millions of charged particles collide each second, a precise description turns out to be inaccessible.

Scientists from HSE and the Yandex Data Analysis School could accelerate the simulation using Generative Adversarial Networks. These include two neural networks that compete with each other at the time of competitive training. This training technique is used, for instance, to create snapshots of people who don’t exist. One network masters how to develop images similar to reality, and the other tries to find differences between real and artificial representations.

It’s amazing how methods that were developed basically to generate realistic photos of cats, allow us to speed up physical calculations by several orders of magnitude.

Nikita Kaseev, Study Coauthor, PhD student, HSE.

Generative competitive networks were trained by the researchers to estimate the behavior of charged elementary particles. The outcomes indicated that it is possible to describe physical phenomena with high precision by using neural networks.

Using generative competitive networks to quickly simulate detector behavior will certainly help future experiments. Essentially, we used the most modern training methods available in data science and our knowledge of the physics of detectors. The diversity of our team, which consisted of data scientists and physicist, also made it possible.

Denis Derkach, Study Coauthor, Assistant Professor, Faculty of Computer Science, HSE.

The research was carried out through funding from the Russian Science Foundation under grant agreement #17-72- 20127 titled “In Search of a New Physics in LHCb Data with the Use of Deep Training Methods.”

Tell Us What You Think

Do you have a review, update or anything you would like to add to this news story?

Leave your feedback