HSE and Yandex Propose Method to Speed Up Neural Networks for Image Generation
A team of scientists at HSE FCS and Yandex Research has proposed a method that reduces computational costs and accelerates text-to-image generation in diffusion models without compromising quality. These models currently set the standard for text-to-image generation, but their use is limited by high computational loads, the company said in a statement.
It is further specified that the developed method—Scale-wise Distillation of Diffusion Models (SwD)—avoids redundant computations during image generation, allowing results to be produced in just 0.3–0.4 seconds.
According to one of the authors, the generation process in diffusion models typically requires dozens of steps involving high-resolution computations. However, in the early stages, only the general structure of the image is formed, and fine details are not yet distinguishable; as a result, some of these computations are redundant. The proposed SwD framework addresses this problem in two ways. First, generation begins at a low resolution and is progressively refined as noise is reduced, eliminating redundant computations in the early stages. Second, the method uses distillation of pre-trained models such as FLUX and Stable Diffusion 3.5, whereby a simpler student model learns to replicate the output of a more complex one, reducing the number of generation steps from dozens to just 4–6.
The authors propose a new loss function for training—Maximum Mean Discrepancy (MMD), which compares how the teacher model 'sees' an image at its internal processing levels with how the student model represents the same image. Unlike traditional approaches, this method does not require auxiliary models, simplifying and accelerating training, the company emphasises. Moreover, MMD can be used as a standalone distillation (knowledge compression) technique: in experiments, the time per training iteration was reduced sevenfold compared to more complex combined approaches.
The new approach reduces generation time from several seconds to just 0.3–0.4 seconds while maintaining visual quality. As a result, SwD makes modern diffusion models faster and more cost-efficient to use, improving their accessibility for practical applications, the company statement says.
The solution is described in a paper to be presented at ICLR 2026, one of the leading conferences on artificial intelligence.
See also:
HSE University to Develop Predictive Analytics System for Icebreaker Motors
Industrial automation is one of the key applications of artificial intelligence. A predictive analytics system for large electric motors is among the solutions being developed for the industry as part of HSE University’s Strategic Technological Project ‘Multi-Agent Platform of AI Solutions for Industry-Specific Tasks.’ What is predictive analytics, how can it improve the operation of electric motors, and what specialists joined forces to develop this technology? Anton Zarubin, Dean of the School of Computer Science, Physics, and Technology at HSE University–St Petersburg and the project development coordinator, explains in this interview with the HSE News Service.
How to Assess Students’ Knowledge in the Age of AI
A researcher at HSE University has proposed a flowchart to help lecturers decide how to assess students who use artificial intelligence. It shows where the use of AI should be restricted and where it can be incorporated into the learning process. The article has been published in IT Professional.
5th Fall into ML Conference to Bring Together Leading AI Researchers
The AI and Digital Science Institute at the HSE Faculty of Computer Science invites researchers, developers and everyone shaping the future of technology to the fifth, anniversary edition of the Fall into Machine Learning conference (Fall into ML 2026). The event will take place on October 23–24, 2026, at the HSE Cultural Centre in Moscow and will become the key meeting point for Russia’s AI community.
HSE University Expands Cooperation with Malaysia in Technology Foresight
HSE University researchers will take part in a study of the future of engineering education in Malaysia, while the Malaysian Industry-Government Group for High Technology (MIGHT) will use the iFORA big-data analysis system to validate the findings of its foresight research. These are the outcomes of a visit by HSE representatives to Kuala Lumpur.
Scientists Train Neural Network to Generate Process Plans from 3D Models
Researchers at the HSE FCS AI and Digital Science Institute have developed CAD2TechSpec, a framework that converts 3D models of mechanical parts into machining process plans—step-by-step instructions for machine tools. The solution aims to reduce the time required for the design and preparation of technical process documentation in mechanical engineering, aircraft manufacturing, and other high-tech industries. The study findings have been published in PeerJ Computer Science.
Researchers Assess Contributions of BRICS Countries to Leading ML/AI Conferences
The HSE Scientometrics Centre analysed more than 104,000 papers published between 2020 and 2025 and presented at ten top-level A* conferences ranked by ICORE 2026. The researchers examined the contributions of Brazil, Russia, India, China and South Africa. Together, these countries accounted for around 40,800 publications, or 39.2% of the total. However, the distribution is highly uneven, and a shared BRICS research space has yet to emerge.
Biologists Discover 'Molecular Fingerprint' of Preeclampsia
Researchers at HSE University employed a new method to model hypoxia in placental cells during pregnancies complicated by preeclampsia and identified molecular markers of tissue hypoxia. Since hypoxia is one of the key mechanisms underlying preeclampsia, these findings are important for a more accurate and timely diagnosis of the disease and for the development of effective treatment methods. The paper has been published in Placenta.
Laboratory of Future Networks: HSE Telecommunications Research Institute Develops 5G/6G Research Testbed
The 5G/6G testbed at the HSE Telecommunications Research Institute is becoming a research platform, an educational laboratory, and a foundation for developing new software components for future networks. It makes it possible not only to observe how a mobile network operates, but also to change its operating conditions and measure the results: data-transmission speed, latency, errors, radio-resource utilisation, and other parameters. Based on the testbed, researchers plan to develop MIMO, O-RAN, xApp, and IAB technologies, as well as experiment with artificial intelligence.
‘Hedgehog’ Versus ‘Relatives’: Researchers Measure How the Brain Responds to Unexpected Words During Natural Speech
Russian neurophysiologists, including researchers from HSE University, have demonstrated the feasibility of using event-related fields (ERFs) to study brain activity during natural speech perception. The researchers showed that this approach can be applied not only to individual words but also to continuous speech. Their findings indicate that words whose meanings differ significantly from the preceding context require longer processing times. The study also reveals that the brain processes function words in two stages: first, it identifies their grammatical role and then uses this information to predict the next word. The study has been published in Frontiers in Human Neuroscience.
HSE Researchers Create New Corpus of Early Child Speech in Russian
Researchers at the HSE Centre for Language and Brain have presented RusLan-M, an open multimedia corpus that makes it possible to trace the development of early child speech in Russian from first words to the emergence of complex grammatical constructions. The database contains around 41 hours of video recordings and more than 35,000 child utterances. The new resource will help researchers study more precisely how children acquire Russian and, in the longer term, develop more reliable tools for assessing speech development. The study has been published in Language Resources and Evaluation.