From Neural Networks to Stock Markets: Advancing Computer Science Research at HSE University in Nizhny Novgorod

© iStock

The International Laboratory of Algorithms and Technologies for Network Analysis (LATNA), established in 2011 at HSE University in Nizhny Novgorod, conducts a wide range of fundamental and applied research, including joint projects with large companies: Sberbank, Yandex, and other leaders of the IT industry. The methods developed by the university's researchers not only enrich science, but also make it possible to improve the work of transport companies and conduct medical and genetic research more successfully. HSE News Service discussed work of the laboratory with its head, Professor Valery Kalyagin.

— Could you please tell us how the laboratory was created?

— It was established in 2011 under the Russian government's mega-grant programme. At that time, the participation of a foreign scientist was a prerequisite for entering the competition. Fortunately, Professor Panagiotis Pardalos from the University of Florida accepted our offer of cooperation. To this day he actively works with HSE University and remains the academic supervisor of the laboratory. Oleg Kozyrev, Eduard Babkin, and Boris Goldengorin all contributed significantly to the preparation of our application. Boris Goldengorin played an important role in establishing the laboratory.

Back then, the study of algorithms for analysing network structures and what is now known as computer science was a new direction for HSE University in Nizhny Novgorod.

Valery Kalyagin

© HSE University

Three years later, their work under the grant was highly praised by the Ministry of Education and Science of Russia, and thus, was extended for two years. When it was over, we applied for the establishment of an international laboratory at HSE University; we received support, and now continue our activities as HSE University's laboratory.

In the early years of our work, we recruited many young researchers who later became well-known scientists and practitioners.

— What interested them about the new laboratory?

— They now have a unique opportunity to work with renowned scientists and develop in a creative environment. Almost everyone who has taken advantage of this opportunity has grown as a scientist, researcher, and educator over the years. From the beginning, the strategy of development has been based on the mandatory combination of research and teaching. All of our researchers now teach. This aspect of their work, the transfer of knowledge and skills, is essential for a scientist's success.

— What has been achieved in this time?

— Over the years, our laboratory has established itself as a well-known scientific centre in Russia and around the world. This is largely due to the efforts of Professor Pardalos, who places a great emphasis on recognition. We maintain numerous connections with colleagues from various universities and research centres. Our laboratory is the co-organizer of a major international conference on optimisation and its applications. We participate in the conference's programme committee, with our supervisor as the multiple honorary chair of the committee.

We actively collaborate with Russian leading institutions, such as MIPT and Moscow State University. We also work closely with the Keldysh Institute of Applied Mathematics of the Russian Academy of Sciences and scientific centres in Siberia and the Urals, including Novosibirsk, Irkutsk, and Yekaterinburg.

— What key research areas are addressed in your works?

—These are mainly computer science topics: network models, technologies for analysing network structures, various optimisation aspects, including combinatorial and discrete optimisation problems on graphs, and their applications to data mining.

— How can this be explained to someone who is not familiar with higher mathematics?

— I will try to explain this in a way that is easier to understand. A network is a group of nodes and connections between them. The most common examples are social networks and telecommunications networks, where the nodes are people or users of a mobile network, and the connections are the communications between them. These can be represented as graphs with special properties or hypergraphs.

The goal of optimisation is also clear—for example, in a social network, one wants to know which nodes to target with information so that it spreads faster, or which nodes to avoid so that a fake message does not spread.

Another type of task that employees work on is searching large databases for information. This is known as the 'nearest neighbour' search problem, where you have a large dataset and a query, and you need to find the closest match to the query in the database.

If the database contains 10–20 items, there will be no significant difficulties. However, when there are a large number of items, it is important to organise the search process quickly and efficiently. To do this, a specialised graph structure is built on the data, which significantly speeds up the search using specialised algorithms.

— Is it possible to use your results in biology or medicine?

— We are researching a class of network models that include some biological networks, such as the network of brain neurons or the network of gene co-expression.

There are billions of neurons in the brain, and we cannot measure anything in these complex networks. However, with the help of electroencephalography (EEG), it is possible to track the activity of individual brain regions and analyse the connections between them. This allows us to create interesting network structures that can be used to study brain activity and diseases, such as Parkinson's disease and epilepsy, which helps to research these conditions.

© iStock

A gene co-expression network (GCN) is constructed based on gene expression profiles for multiple samples or experimental conditions. Researchers seek to identify pairs of genes with similar expression patterns in all samples, leading to the creation of a network model. This analysis is used for various practical purposes, such as highlighting the most important nodes in the model. The identified cluster of genes suggests that the gene and its neighbours have similar expression patterns, which further simplifies drug testing processes.

— How applicable are your works in the field of economics?

— Another well-known type of network is the stock market. We analyse assets and identify connections between them, which contributes to the formation of a stock market network. The analysis of these networks allows us to create investment portfolios, such as the classic Markowitz optimal portfolio model. While using these models, it is important to remember that they don’t guarantee a risk-free investment, as there is always a chance of loss.

Major trading companies, banks, and investment advisors are looking for ways to construct portfolios that minimise risk while maximising returns. They desire stability and security in their investments, and network models can help them achieve these goals. By understanding the relationships between assets, they can identify portfolios that best meet their criteria.

— You and your colleagues must be incredibly rich.

— We do not engage in trading or provide investment advice. Students write graduation papers and research on these and related topics, analysing how different portfolios perform in various market conditions.

While this is not a substitute for professional analysis, it can be a useful starting point for exploring additional opportunities in the stock market.

One example of how this can be applied is by selecting a portfolio through the creation of a network market graph, where independent sets are identified. Experimental evidence suggests that such sets can lead to diverse and profitable portfolios.

— Do the models you have created suggest different development scenarios?

— The laboratory is actively exploring the uncertainty of algorithms used to construct various graph structures in network models, such as gene co-expression networks, brain networks, and stock market networks.

If the uncertainty is high, it is possible that the conclusions drawn from these models may be false. We always hope to get rich, but our expectations may not be met.

— How does the solution of fundamental scientific problems combine with applied work?

— We have a strong team led by Dmitriy Malyshev. The group’s focus is on algorithmic graph theory, which is closer to theoretical computer science and discrete mathematics. A significant number of graduate students and young laboratory staff have defended their theses on these topics. Despite the fundamental theoretical nature of the research, it also has applied significance. Estimating the computational complexity of graph problems help us to identify computationally difficult problems and find classes of problems that can be solved efficiently.

© iStock

In the early years of the laboratory's work, we developed the field of data mining and AI. It is led by Andrey Savchenko. He develops the field of data mining in conditions of limited resources, such as on mobile devices that are less powerful than desktop computers or laptops. For example, we want to classify photos, texts, or something else on our smartphones, but due to the limitations of these devices, it is impossible to deploy full-fledged neural networks. However, Andrey Savchenko and his team developed an approach that allowed them to effectively tackle these challenges. As a result, they patented this approach as their intellectual property. Today, there are apps available that can be downloaded and utilized, making use of this new technology.

— Do we need this now, when we are promised quantum computers with unlimited possibilities?

— Head of the research centre of a large foreign company said recently that we have returned to the situation of the 1970s, when scientists and practitioners, given the limited capabilities of processors and computer memory, paid special attention to the effectiveness of algorithms. Then the processor speed and memory capacity, including RAM, increased dramatically, and this somewhat lost its relevance. Now the problem has returned, as we do not expect major improvements in the hardware. When you train large language models or search large databases, you return to the need for fast calculations in conditions of limited resources. Currently, many large manufacturers of computing resources and IT companies are conducting research on the effective use of existing capabilities. If we reduce the calculations on at least one node by 1%, we will see a significant effect. We were successful in a project with an IT company where we applied computational graph patterns to accelerate the learning of neural networks. Such tasks are becoming more and more popular.

The emergence of a quantum computer with unlimited possibilities is still not a matter of the very near future.

— Which companies have used your developments?

— We have developed an algorithm to organise the delivery of products to stores for a large retail chain. This task is called the transport routing problem, which is a network problem that calculates traffic on a network of roads. It has a high level of computational complexity. If there are 100 cars and 1,000 stores, and we want to optimise traffic, it would be difficult to solve the problem manually. It is also difficult for a computer to solve, but smart algorithms can help. This makes it possible for AI to manage the logistics of using transport.

— Is there a problem with the outflow of scientists to industry partners?

— There is a challenge with staff leaving for IT companies. When we start interacting with companies, they see the qualifications of our staff, invite them to work on science projects and solve interesting problems, and then headhunt specialists with better offers.

— Which departments of HSE University does the laboratory cooperate with?

— The closest cooperation has been established with the International Centre of Decision Choice and Analysis and with the Laboratory for Applied Network Research.

— What perspective do you see for research?

— We focus on a combination of fundamental and applied research in order to produce both good theoretical results and publications, as well as collaborate with industry on joint projects.

The university's strategy is to increase its emphasis on applied research, which is a trend across Russia. We aim to learn how we can meaningfully contribute to the country's economic and social growth through our theoretical developments. We see potential in developing algorithms and technologies for artificial intelligence systems.

Alongside the purely scientific component, we also aim to popularise science by making theoretical and applied findings accessible to schoolchildren, as they are our future students and lab staff.

As a leading scientific centre in computer science and its application, the laboratory welcomes new partnerships for both fundamental and applied projects.

— What educational programmes do you take part in?

— We are involved in two key programmes on campus: Applied Mathematics and Information Science (Bachelor's degree) and Data Mining (Master's degree). These programmes incorporate laboratory work, which is an integral part of both the teaching process and the research conducted by our students.

All international laboratories develop research expertise and pass it on to young people. If we have no contact with students, where will we recruit new staff?

I would like to add that our graduates are in demand in many companies and countries.

— Why is it important to preserve fundamental research?

— Currently, we see the rebirth of mathematics. The development of data mining and artificial intelligence technologies has created tasks that require specialists with developed abstract thinking and a broad outlook, which fundamental mathematics provide. At that, multiple branches of mathematics are in demand. This is a sign of the 21st century.

For example, we have a huge dataset and are trying to understand how it works. Often, the high dimension of data is an obstacle in their analysis. To reduce the dimension without losing information, it is necessary to understand many sections of fundamental mathematics well, from classical methods of linear algebra and mathematical analysis to advanced probabilistic models and topology.

Mathematicians are getting excited, people see that they need to expand their field of activity to applied research, which is a characteristic feature of HSE University.

— How do you manage to maintain international relations?

— We continue to maintain contact with foreign scientists. Since 2012, we have been holding an annual international conference on network analysis and international schools for young scientists. Almost everyone who comes to Nizhny Novgorod continues to communicate and responds to our offers, despite the pandemic and current situation. These events provide young scientists with an additional opportunity to assess the level of their research, and it becomes even clearer when they interact with colleagues from abroad. We encourage young people to engage actively with the guests, and students also show interest in this process.