HSE MIEM Student on Internship at CERN

Elena Orlova, a fourth year student in Applied Mathematics at MIEM, took part in a summer school at CERN (European Organization for Nuclear Research, Geneva, Switzerland). As part of the GeantV Software Developer Team, she relied on machine learning to model the transition of elementary particles through matter.

This summer, CERN hosted an annual school, which attracts students from all over the globe. Elena Orlova, a student of Applied Mathematics at HSE MIEM School of Applied Mathematics, took part in the school. Over the course of nine weeks, the students attended various lectures and worked on practical research projects under the supervision of CERN scholars. In addition to studies, they took part in various informal events, such as hackatons, as well as excursions to Zurich, Google, OpenSystems, ETH Zurich and EPFL Lausanne.

Elena shares her story about working at CERN below.

‘It all started with an interdisciplinary term paper that I wrote as a third year student - “Calculation of an Electron Spectrum in a Spherically Symmetric Potential”. This was dedicated to computational methods in quantum physics and was supervised by Renat Ikhsanov, Associate Professor at the HSE MIEM School of Electronic Engineering. Computational methods in physics have turned out to be a quite thrilling area, which I wanted to continue investigating. So, I applied for the summer school at CERN. I filled in the online application and asked my academic supervisor to provide me with a recommendation letter.

‘As part of the project at CERN, I modeled the transition of elementary particles through matter and calculated showers – cascades of secondary particles, which are produced as the result of particles’ interactions with a layer of matter in a detector. Modeling is an important component in analyzing the transition of particles through matter, from designing the detector, to a final comparison of the experimental data with theoretical models. Furthermore, Monte Carlo methods, which require large computational resources, are usually used to model the transition of elementary particles through matter. However, in order to process the data from next-generation experiments on the Large Hadron Collider (LHC), even the substantiate CERN computational resources may not be enough. Therefore, scholars are now trying to develop new methods of modeling, with machine learning being one of them.

‘As part of my project, I utilized the Generative Adversarial Networks (GANs) model, which is one of the freshest and most promising approaches to Computer Vision and Deep Learning. This approach is based on the following process. We have two neural networks. The first, called a generator, generates images from a random noise vector. The second, called a discriminator, categorizes the images as “real” or “false”. The key advantage of this approach is that it allows for modeling of any type of elementary particles’ detector.

‘I worked with the developer team of the GeantV software system, which is a global leader in calculating the transition of elementary particles through matter. In addition, the team includes many researchers from Russia. I conducted my project relying on the Keras and Neon frameworks, and also tested the program using various CPU and GPU clusters.

‘The project relied on real data from detectors installed on the LHC. In addition to data on particle energy, our model can generate information on the type of particle and its initial energy. The results we got appear to be promising, and the GeantV Developer Team hopes that the project will help physicists process data from new experiments conducted on the LHC.’