💥Drumroll, please: Skoltech and Yandex are launching a new master’s program, "AI Robotics"!
The program trains robotics engineers who develop intelligent systems at the intersection of software and hardware. It is designed for graduates of technical fields and developers who want to dive deeper into robotics, AI, and industrial automation.
Graduates will also be able to become in-demand intelligent robotics and process automation engineers, R&D project managers who know how to manage the development of new products: from research to industrial implementation, coordinating teams of engineers and analysts. The training is conducted entirely in English.
👉 Check out this new master’s program at our Open Doors Day this Thursday — register today, spots are filling up fast!
See you at Skoltech!
The program trains robotics engineers who develop intelligent systems at the intersection of software and hardware. It is designed for graduates of technical fields and developers who want to dive deeper into robotics, AI, and industrial automation.
Graduates will also be able to become in-demand intelligent robotics and process automation engineers, R&D project managers who know how to manage the development of new products: from research to industrial implementation, coordinating teams of engineers and analysts. The training is conducted entirely in English.
👉 Check out this new master’s program at our Open Doors Day this Thursday — register today, spots are filling up fast!
See you at Skoltech!
‼️Heads up
Due to high demand, registration for the Skoltech Summer School on Machine Learning has been extended.
Submit your application by 11:59 p.m. April 23.
The SMILES-2025 will take place from July 14 to July 27 at the Harbin Institute of Technology (HIT). This year’s main theme is generative methods in AI. The program will focus on such topics as large language models, multi-agent systems, AI safety, and generative approaches in computer vision.
The summer school is co-organized by Skoltech, the HIT Faculty of Computer Science, with Alfa Bank as the main partner and AIRI Institute as the scientific partner.
👉 More details here.
Due to high demand, registration for the Skoltech Summer School on Machine Learning has been extended.
Submit your application by 11:59 p.m. April 23.
The SMILES-2025 will take place from July 14 to July 27 at the Harbin Institute of Technology (HIT). This year’s main theme is generative methods in AI. The program will focus on such topics as large language models, multi-agent systems, AI safety, and generative approaches in computer vision.
The summer school is co-organized by Skoltech, the HIT Faculty of Computer Science, with Alfa Bank as the main partner and AIRI Institute as the scientific partner.
👉 More details here.
📸 Here's the photo album from the recent Skoltech's Open Doors!
During this event, we presented our educational programs, detailed the application and admission process, and showcased our campus and laboratories.
Don’t miss our next Open Doors event on April 24, where we will discuss the fascinating field of intelligent robotics!
👉 Register now to secure your spot!
During this event, we presented our educational programs, detailed the application and admission process, and showcased our campus and laboratories.
Don’t miss our next Open Doors event on April 24, where we will discuss the fascinating field of intelligent robotics!
👉 Register now to secure your spot!
⭐️ Skoltech recently hosted the annual Career Day, designed to help students take their first steps toward their professional journey.
This year, the Skoltech Career Center, together with its partners, organized an intensive program for students:
- A career opportunities exhibition
- Individual career consultations with HR specialists and hiring managers, offering guidance on CVs, insights about companies, industry-specific labor market trends, and assistance in creating personalized career paths
- "Speed dating" with graduates
- Quizzes from partners and branded prizes
The event attracted participants from over 20 companies, including the Russian Research Institute, Kamaz, Sber, BIOCAD, Yandex, Safe Transport, Cyber Motorika, T-Bank, Wildberries, OilGene, RKC, Virsign Innovations, Sibur Polilab, Piklema, Start in Moscow, NGO Varya, KEPT, M&S Decisions, NSPC (National Payment Card System), Rosselkhozbank, and MTS Digital.
This year, the Skoltech Career Center, together with its partners, organized an intensive program for students:
- A career opportunities exhibition
- Individual career consultations with HR specialists and hiring managers, offering guidance on CVs, insights about companies, industry-specific labor market trends, and assistance in creating personalized career paths
- "Speed dating" with graduates
- Quizzes from partners and branded prizes
The event attracted participants from over 20 companies, including the Russian Research Institute, Kamaz, Sber, BIOCAD, Yandex, Safe Transport, Cyber Motorika, T-Bank, Wildberries, OilGene, RKC, Virsign Innovations, Sibur Polilab, Piklema, Start in Moscow, NGO Varya, KEPT, M&S Decisions, NSPC (National Payment Card System), Rosselkhozbank, and MTS Digital.
🏆 A Skoltech student received an award for the best scientific paper
Peter Sychev, a master’s student in the Data Science program, has received the Best Paper Award at AINL 2025, an international conference on artificial intelligence and natural language processing held in Novosibirsk.
📝 Paper Title: "When an LLM is apprehensive about its answers — and when its uncertainty is justified"
🔗 arXiv: 2503.01688
What’s the research about?
The researchers tested whether large language models (like ChatGPT) can recognize when they’re wrong. To do this, they compared two different methods of assessing a model’s "confidence" in its answers.
In the first case, they looked at how evenly the model distributes probabilities across different answer options — if all choices seem almost equally likely, then it probably doesn't know the right one.
In the second, the model itself evaluates how well it answered.
It turned out that the first method works well for simple facts, but the second method helps only in easy tasks, and in more complex tasks it often "does not notice" errors.
The study also showed that even popular AI tests can hide biases, and reminded that when creating reliable AI systems, it is important to take into account not only the errors of the model itself, but also the peculiarities of the data on which it was trained.
🎉 Congratulations to Peter on his well-deserved award! Wishing him future scientific achievements!
Peter Sychev, a master’s student in the Data Science program, has received the Best Paper Award at AINL 2025, an international conference on artificial intelligence and natural language processing held in Novosibirsk.
📝 Paper Title: "When an LLM is apprehensive about its answers — and when its uncertainty is justified"
🔗 arXiv: 2503.01688
What’s the research about?
The researchers tested whether large language models (like ChatGPT) can recognize when they’re wrong. To do this, they compared two different methods of assessing a model’s "confidence" in its answers.
In the first case, they looked at how evenly the model distributes probabilities across different answer options — if all choices seem almost equally likely, then it probably doesn't know the right one.
In the second, the model itself evaluates how well it answered.
It turned out that the first method works well for simple facts, but the second method helps only in easy tasks, and in more complex tasks it often "does not notice" errors.
The study also showed that even popular AI tests can hide biases, and reminded that when creating reliable AI systems, it is important to take into account not only the errors of the model itself, but also the peculiarities of the data on which it was trained.
🎉 Congratulations to Peter on his well-deserved award! Wishing him future scientific achievements!
👨💻 Scientists from Skoltech and MIPT together with Russian and foreign colleagues have developed a new method for parametrizing machine-learning interatomic potentials (MLIP) to simulate magnetic materials. It significantly increases the prediction of their properties and makes it much more reliable and accurate. The research was published in Computational Materials Science and supported by the Russian Science Foundation.
The authors were able to create accurate and reliable MLIPs that require a limited amount of expensive training data. Their key idea was to train MLIPs on magnetic forces — negative derivatives of energy with respect to magnetic moments — in addition to energies, atomic forces, and stresses. The training was performed on data computed for about 2,600 different atomic configurations of an iron-aluminum (Fe-Al) alloy with different component ratios.
The new method showed a tenfold reduction in prediction error for magnetic forces, but virtually no change for energies and conventional forces. The models trained on magnetic forces also proved to be more accurate in predicting the equilibrium magnetic moments of iron atoms.
The authors were able to create accurate and reliable MLIPs that require a limited amount of expensive training data. Their key idea was to train MLIPs on magnetic forces — negative derivatives of energy with respect to magnetic moments — in addition to energies, atomic forces, and stresses. The training was performed on data computed for about 2,600 different atomic configurations of an iron-aluminum (Fe-Al) alloy with different component ratios.
The new method showed a tenfold reduction in prediction error for magnetic forces, but virtually no change for energies and conventional forces. The models trained on magnetic forces also proved to be more accurate in predicting the equilibrium magnetic moments of iron atoms.