HSE Opens Laboratory of Financial Data Analysis

© Essentials/ iStock

Part of the Centre of Deep Learning and Bayesian Methods and another partner project between Sberbank and HSE University’s Faculty of Computer Science, the laboratory will focus on applying machine learning methods to financial services.

The lab’s research agenda includes the interpretation of complex neural network models, reinforcement learning, natural language processing, and competing networks (GAN) in directional information removal from samples. The laboratory will be headed by Evgeny Sokolov, Deputy Head of the Big Data and Information Retrieval School at HSE. According to Sokolov, the idea of ​​creating a new research unit was a natural continuation of joint projects with Sberbank—a bank HSE has been collaborating with for years.

 
Evgeny Sokolov

There is a great need for specialized research, insofar as methods—such as credit risk assessment, support service automation, and marketing personalization—often require refinement when applied to banking.

Now, large banks are introducing machine learning more and more, both for basic tasks and for new areas such as creating chat bots

One of the most common areas machine learning is applied to is credit scoring. A computer model processes data about people who have already paid a loan—their gender, age, marital status, income level—and finds patterns in them. The combination of these factors is then used to determine how much the bank risks when issuing a loan.

But it is not enough to teach the neural network to qualitatively predict the borrower's creditworthiness based on the available data. The bank must be able to justify its final decision—this is a requirement of the Central Bank of the Russian Federation, which regulates credit institutions. Due to the difficulties in interpreting the predictions of complex neural networks, simplified, less accurate models are still used in practice.

If banks were allowed to use neural networks, the quality of credit risk assessment would increase dramatically

According to Sokolov, the lab plans to recruit students from the Master’s Programme ‘Financial Technologies and Data Analysis’, which was launched jointly with Sberbank in 2017, to work as interns in the lab. Interns will be able to participate in developing solutions to challenges in machine learning with the guidance of experienced researchers, as well as get to know the inner workings of the bank and interact with its developers and data scientists.

Part of the laboratory research will involve not only banking but other financial spheres. A major business trend is automatizing customer support services. Every day customer service call centers receive tens of thousands of calls from clients with problems that are similar to one another. About 80% of responses to these requests can be provided by a template, and a chat bot can handle them. Once you accumulate enough data, you can begin to automate processes by using machine learning and natural language processing.