Alexey Zaytsev, senior lecturer at Skoltech and head of the LARSS laboratory, along with Saveliy Galochkin, a research engineer at the Applied AI Center, took part in the New League of Universities conference held on the Skoltech campus. Besides our university, the association of universities includes RANEPA, ShaniNKA, and the European University in St. Petersburg. The conference's objectives included discussing potential joint projects, both academic and applied.
In his first presentation, Alexey Zaytsev focused on assessing the unreliability of neural networks and the associated risks and opportunities. He elaborated on how neural networks function, how one can obtain a neural network in an explicit form for simple models, and how to approximate it for more complex neural network models. He also presented Skoltech's latest work in this area, including uncertainty estimation in face recognition tasks and for language models of various sizes. In his second presentation, Alexey discussed predicting climate risks using deep learning technologies, where predictive models help forecast dangerous storm winds, hail, floods, and droughts with unprecedented accuracy.
Saveliy Galochkin discussed the application of AI and computer vision technologies for preserving cultural heritage objects using the example of the Cathedral of Vasily the Blessed. The project aimed to create a complete 3D model of the cathedral, for which an installation with various types of cameras was used for highly accurate and detailed shooting of small objects, as well as a laser scanner and a special camera for scanning details with textured surfaces in conditions of limited lighting, which cannot be intensified due to the potential harm of bright light to the monument. The scanner was used to capture the geometry of objects, and for their more detailed reconstruction and further texture overlay, a photogrammetric camera was utilized. To scan the cathedral domes, the team deployed drones equipped with sensors.