HSE University Students Win in the AIJ Science Competition at AI Journey 2023
The International Sber Conference of Artificial Intelligence, ‘AI Journey 2023’ recently took place in Moscow. Alexander Rogachev, doctoral student of the HSE Faculty of Computer Science, and Egor Egorov, an HSE 4th-year undergraduate student became the winners of the AIJ Science competition for scientific articles on artificial intelligence that was held as part of the event. The research was carried out under the umbrella of the HSE's Laboratory of Methods for Big Data Analysis (LAMBDA).
Every year artificial intelligence is becoming more and more integrated into various areas of human life. New tasks that require the use of algorithms to solve specialised problems emerge, and new architectures and approaches designed to bring people closer to creating more efficient technologies are being developed. The first day of AI Journey 23 was dedicated to advanced scientific research, key developments of the year and trends in the field of artificial intelligence and machine learning.
An open selection of scientific articles was held as part of the conference. HSE University students presented an article dedicated to the influence of adaptive spectral normalisation on the quality of generative models and the stability of their learning. The researchers analysed approaches that allow us to find a balance between the model’s stability and expressiveness. This study is based on data from the CERN Large Hadron Collider. However, the authors note that these results can be applied to other data and models — for example, for image generation.
The article will be published in a special issue of the Doklady Mathematics journal, which is being published to support research activities in the field of data analysis, artificial intelligence and machine learning, as well as to exchange opinions and practical experience. The journal belongs to the Q1 category.
Doctoral student and lecturer at the Big Data and Information Retrieval School at HSE Faculty of Computer Science
‘In our research we studied the possibility of using adaptive methods of regularisation of generative adversarial networks (GAN) based on spectral normalisation. As we know, GAN can have stability problems during the learning process, and spectral normalisation is often used to solve these problems. However, stability correlates with the expressiveness of the model. We analysed the possibility of a balance between these two characteristics through the example of a problem simulating a physical process within the LHCb experiments using a generative network.’
Egor Egorov
4th-year student of the HSE programmme in Applied Mathematics and Information Science
‘During our work on this project, we obtained some interesting results and decided to formalise them in the form of an article. I am glad that experts in the field of ML selected and appreciated our research. This motivates us to keep on working and developing in our academic research.’