Добро пожаловать на регулярные семинары по актуальным темам исследований в области вычислительной механики!

Приглашённые лекторы из Сколтеха и других вузов выступают с докладами, чтобы познакомить студентов с текущими исследованиями и достижениями в различных областях современной механики жидкости и твёрдого тела, прикладной математики, вычислительной математики и промышленного применения механики. Студенты получают возможность узнать об актуальных проблемах механики у ведущих специалистов в области вычислительной механики.

Продолжительность доклада: 50 минут

Q&A: 10-15 минут

Семинары проводятся на английском языке.

Ведущий преподаватель: Аслан Касимов, доцент

Контакты: A.Kasimov@skoltech.ru

2 АПРЕЛЯ, 14:00 | НОВЫЕ КАТАЛИЗАТОРЫ ДЛЯ УСТОЙЧИВОГО РАЗВИТИЯ И ДЕКАРБОНИЗАЦИИ

Аудитория: R3-2009 
Докладчик: Александра Радина, аспирантка программы «Науки о материалах», Сколтех

Nowadays the most widely used catalysts for organic synthesis are mainly made of noble and rare earth metals which significantly increases the cost of many products. Moreover, 80% of catalyst used in domestic industry including oil and chemical production are imported, and it can cause problems in the present political situation. To solve this problem, the Industry-Oriented Computational Discovery Group at Skoltech is currently conducting studies to find a new catalyst that would be as effective as those made from precious metals, but cheaper. Transition metal carbides and borides have been suggested as promising candidates for new catalysts. Recently, some of our theoretical predictions have been confirmed experimentally. It has been shown that higher transition metal borides have great catalytic activity for the conversion of CO2 into various fuels and hydrogen production processes.

9 АПРЕЛЯ, 14:00 | ИССЛЕДОВАНИЕ МЕХАНИЗМОВ ФОРМИРОВАНИЯ ПРОСТРАНСТВЕННО-ВРЕМЕННЫХ СТРУКТУР, ВОЗНИКАЮЩИХ НА ДВИЖУЩЕМСЯ ФРОНТЕ РЕАКЦИИ

Аудитория: R3-2009 
Докладчик: Эдуард Якупов, Младший научный сотрудник, Лаборатория нелинейной динамики и теоретической биофизики, Физический институт имени П. Н. Лебедева Российской академии наук

In experimental studies of combustion wave propagation in gaseous media, it has been found that under certain conditions autowaves - spirals or targets - patterns appear at the wave front. This study aims to solve pressing problems in nonlinear dynamics by examining autowave and dissipative structures emerging at a propagating reaction front. It explores the formation and evolution of complex spatiotemporal structures in distributed dynamic systems, focusing on processes occurring at the reaction front. The research employs an approach that considers the hierarchical nature of spatial-temporal self-organization via block models. Key outcomes include developing a novel methodology for analyzing the mechanisms underlying these structures' formation, detailing conditions necessary for the emergence of autowave and Turing structures, creating a reduced model for high-pressure hydrogen-air combustion, and establishing quantitative criteria for the formation of different structure types based on system parameters.

16 АПРЕЛЯ, 14:00 | ПРОБЛЕМА СОГЛАСОВАНИЯ МОДЕЛЕЙ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА С ЭМПИРИЧЕСКИМИ ДАННЫМИ ПРИ ГИБРИДНОМ МОДЕЛИРОВАНИИ

Аудитория: R3-2009
Докладчик: Иван Тюкин, Профессор, Центр искусственного интеллекта, Сколтех

Recent years have seen drastic increase in the popularity of advance machine learning & AI tools such as neural networks in various modelling settings. Nowadays classical solvers and approaches often go hand-in-hand with their data-driven counterparts complementing each other as appropriate. Relevant use-cases include physics-informed neural nets as replacements for solvers of large, complicated problems or modelling complex empirical relationships in otherwise classical models by neural nets. In these and many other similar cases, one of the pressing questions is how much data we need to fit our model to data and what guarantees can be made for a given model and dataset at hand. In this talk we will delve into the problem of fitting models to data with the view to develop an understanding of basic mathematical ingredients needed to answer these questions. We will look at the problem from two alternatives directions: statistical learning theory and approximation theory. We will discuss similarities and discrepancies between these two approaches. We will also present a recent example of the application of machine learning for extrapolating solutions of integro-differential equations. The seminar is intended for a broad audience of PhD students and researchers interested in developing a normative and rigorous take on the application of neural networks in modelling.