HSE Scientists Optimise Training of Generative Flow Networks
Researchers at the HSE Faculty of Computer Science have optimised the training method for generative flow neural networks to handle unstructured tasks, which could make the search for new drugs more efficient. The results of their work were presented at ICLR 2025, one of the world’s leading conferences on machine learning. The paper is available at Arxiv.org.
Generative Flow Networks (GFlowNets) are a class of machine learning algorithms that build complex objects step by step. Researchers use them to search for new proteins and drugs, and to optimise transport systems.
For GFlowNets to discover such complex structures, researchers specify the desired properties of the target object. The closer the network’s proposed solution is to these properties, the higher the reward it receives. GFlowNets aim to solve problems in a way that maximises their reward. They do not rely on data directly, but instead on the reward, which is computed using an equation known as the value function.
The search for a complex object can be compared to assembling a Lego model, where pieces are added step by step until the object is complete, with each model assigned a specific value—for example, a plant model might be valued higher than an animal model. Unlike other machine learning methods that would strive to construct a plant at any cost, GFlowNets generate a variety of objects—but plants more frequently than animals—because the reward for plants is higher.
In this type of search, GFlowNets rely on two stochastic policies that operate together: a forward policy and a backward policy. The forward policy can be thought of as a construction foreman, deciding the next step and estimating the probability of the subsequent state, while the backward policy acts as a deconstruction expert, identifying the preceding step. Maintaining balance between these flows is crucial but difficult to achieve. First, it requires significant computing power. Second, backward policies lack flexibility: researchers usually prevent them from adapting during the search or from observing the actions of the forward policy.
HSE scientists have developed a way to optimise backward policies using a method called Trajectory Likelihood Maximisation (TLM). They refined the backward policy’s algorithms so that it can be continuously checked against the steps of the forward policy.
'We designed the search for the optimal solution to resemble a negotiation, where both sides are ready to adjust their positions. In highly uncertain problems, the backward policy serves only as an auxiliary tool that improves the results of the forward policy. Our goal was to make the backward policy more flexible, and we finally succeeded,' explains Timofey Gritsaev, co-author of the paper and Research Assistant of the Centre for Deep Learning and Bayesian Methods at the HSE FCS AI and Digital Science Institute.
After implementing TLM, the reward function that measures the backward model’s success became more complex. Nevertheless, despite this increased complexity, the overall search system became faster and more efficient.
'Our method explores the space of possible solutions noticeably faster and identifies more high-quality options. Overall, this approach brings generative models closer to reinforcement learning methods,' explains Nikita Morozov, Junior Research Fellow of the Centre for Deep Learning and Bayesian Methods at the AI and Digital Science Institute of the HSE FCS.
The authors of the study are confident that their work will benefit specialists using GFlowNets across various fields, including the search for new medicinal compounds, the development of materials with specific properties, and the fine-tuning of large language models. Thanks to these networks’ ability to efficiently explore vast solution spaces and quickly identify the best options, the demand on computing power can be significantly reduced.
See also:
Scientists Test Asymmetry Between Matter and Antimatter
An international team, including scientists from HSE University, has collected and analysed data from dozens of experiments on charm mixing—the process in which an unstable charm meson oscillates between its particle and antiparticle states. These oscillations were observed only four times per thousand decays, fully consistent with the predictions of the Standard Model. This indicates that no signs of new physics have yet been detected in these processes, and if unknown particles do exist, they are likely too heavy to be observed with current equipment. The paper has been published in Physical Review D.
HSE Scientists Reveal What Drives Public Trust in Science
Researchers at HSE ISSEK have analysed the level of trust in scientific knowledge in Russian society and the factors shaping attitudes and perceptions. It was found that trust in science depends more on everyday experience, social expectations, and the perceived promises of science than on objective knowledge. The article has been published in Universe of Russia.
Institute for Robotics Systems Established at HSE University
As decided by the HSE University Academic Council, a new Institute for Robotics Systems will be established at HSE, and with a strong fundamental base. It will cooperate with relevant departments across the university and engage students and doctoral candidates in research and development (R&D). First Vice Rector of HSE University and Director of the Institute for Statistical Studies and Economics of Knowledge, Leonid Gokhberg, discussed the expected practical results and the framework for cooperation with an industrial partner.
HSE Tops Ranking of Universities Participating in Priority 2030 Programme
The Russian Ministry of Science and Higher Education has published an updated list of participants in the Priority 2030 programme. A total of 106 universities will receive support this year. HSE University was included in the first group and topped the ranking.
HSE Psycholinguists Launch Digital Tool to Spot Dyslexia in Children
Specialists from HSE University's Centre for Language and Brain have introduced LexiMetr, a new digital tool for diagnosing dyslexia in primary school students. This is the first standardised application in Russia that enables fast and reliable assessment of children’s reading skills to identify dyslexia or the risk of developing it. The application is available on the RuStore platform and runs on Android tablets.
Physicists Propose New Mechanism to Enhance Superconductivity with 'Quantum Glue'
A team of researchers, including scientists from HSE MIEM, has demonstrated that defects in a material can enhance, rather than hinder, superconductivity. This occurs through interaction between defective and cleaner regions, which creates a 'quantum glue'—a uniform component that binds distinct superconducting regions into a single network. Calculations confirm that this mechanism could aid in developing superconductors that operate at higher temperatures. The study has been published in Communications Physics.
Neural Network Trained to Predict Crises in Russian Stock Market
Economists from HSE University have developed a neural network model that can predict the onset of a short-term stock market crisis with over 83% accuracy, one day in advance. The model performs well even on complex, imbalanced data and incorporates not only economic indicators but also investor sentiment. The paper by Tamara Teplova, Maksim Fayzulin, and Aleksei Kurkin from the Centre for Financial Research and Data Analytics at the HSE Faculty of Economic Sciences has been published in Socio-Economic Planning Sciences.
Larger Groups of Students Use AI More Effectively in Learning
Researchers at the Institute of Education and the Faculty of Economic Sciences at HSE University have studied what factors determine the success of student group projects when they are completed with the help of artificial intelligence (AI). Their findings suggest that, in addition to the knowledge level of the team members, the size of the group also plays a significant role—the larger it is, the more efficient the process becomes. The study was published in Innovations in Education and Teaching International.
New Models for Studying Diseases: From Petri Dishes to Organs-on-a-Chip
Biologists from HSE University, in collaboration with researchers from the Kulakov National Medical Research Centre for Obstetrics, Gynecology, and Perinatology, have used advanced microfluidic technologies to study preeclampsia—one of the most dangerous pregnancy complications, posing serious risks to the life and health of both mother and child. In a paper published in BioChip Journal, the researchers review modern cellular models—including advanced placenta-on-a-chip technologies—that offer deeper insights into the mechanisms of the disorder and support the development of effective treatments.
Advancing Personalised Therapy for More Effective Cancer Treatment
Researchers from the International Laboratory of Microphysiological Systems at HSE University's Faculty of Biology and Biotechnology are developing methods to reduce tumour cell resistance to drugs and to create more effective, personalised cancer treatments. In this interview with the HSE News Service, Diana Maltseva, Head of the Laboratory, talks about their work.