Skoltech PhD students Ivan Butakov, Ilya Zisman and Artem Lykov, Research Engineer Alexander Kolesov and Assistant Professors Alexander Tyurin and Alexander Korotin have been awarded prestigious international Yandex ML Prizes. The winners will receive cash prizes ranging from 500,000 to 1,000,000 rubles depending on the prize category, access to Yandex 360 services, and a 500,000-ruble grant to use the Yandex Cloud platform.

Engineering Systems PhD student Artem Lykov who pursues research in cognitive robotics, generative AI, and large language models, received the award for his work titled “CognitiveDog: Large Multimodal Model Based System to Translate Vision and Language into Action of Quadruped Robot.” Prepared under the supervision of Associate Professor Dzmitry Tsetserukou and co-authored by researchers from Skoltech’s Laboratory of Intelligent Space Robotics, the paper presents the world’s first intelligent robotic dog capable of communicating with humans, manipulating physical objects, and creating action plans to solve non-trivial tasks.

“I wouldn’t put a date on it, but I think cognitive robots will be available to the general public in about five years,” Lykov said. “They will probably cause as much amazement as generative AI did some time ago. Now that we have both generative AI and physical robots, the only thing preventing their wide use is more reliable tools for connecting them.”

Ivan Butakov, a PhD student in Computational and Data Science and Engineering, was recognized for his research in information theory, learning theory, generative models, and unsupervised learning. Proving the hypothesis that neural networks prefer to forget unnecessary information is one of his most notable achievements.

Ivan’s findings could help control AI’s memory so that it can better remember heterogeneous tasks or forget certain information, such as accidentally exposed personal data. “Sometimes we need to make AI forget certain things,” Butakov explained. “On the other hand, it is equally important to prevent memory lapses and motivate the neural network to remember the information it has learned during training on different types of tasks or datasets.”

Ilya Zisman, a PhD student in Computational and Data Science and Engineering, won a prize for his research in reinforcement learning and meta-learning, “I am happy to receive the award, and even more happy to be part of the community of prize winners! At the awards ceremony, you meet and exchange ideas with new and past winners, which is very important for a scientist.”

Research Engineer Alexander Kolesov from the Applied AI Center received a prize for his research on Wasserstein barycenters and generative models. He emphasized that the search for the Wasserstein barycenter — the averaging point for different images — remained a purely theoretical task for years due to limitations in finding a suitable practical application.

“The newly designed efficient algorithm marks a new milestone in this research area, contributing to both theoretical insights and various practical applications of the algorithm,” Kolesov commented.

Assistant Professor Alexander Tyurin from the Applied AI Center was named among the best researchers for his work on non-convex and convex optimization, stochastic optimization, asynchronous methods, decentralized optimization, and federated learning. “My research addresses problems such as asynchronous optimization, information compression, and random device shutdowns. The award means that my work has been recognized by top researchers and that my efforts have borne fruit, which is very motivating for further research,” Tyurin noted.

Assistant Professor Alexander Korotin from the Applied AI Center was recognized as one of the best research supervisors. His research interests range from generative models, unpaired learning and domain translation to optimal transport, Schrödinger bridges, diffusion models, and adversarial learning. “Generative approaches based on optimal transport are as efficient as other models in solving some data generation problems. They also open new horizons for practical and theoretical research,” Korotin said.

From 160 submissions this year, the jury selected 14 winners representing the most impactful research in generative models, natural language processing, computer vision, information retrieval, speech recognition and synthesis, and cognitive robotics.