About maximal distance minimizers. What is it and why they are so good? Докладчик: Яна Теплицкая Дата: 03.09.2024 г. Час: 14:00 ч. Място: зала 256 Резюме  

Subdiffusion-reaction systems: from microscopic random walk to the mesoscopic fractional PDE's Докладчик: проф. Сергей Федотов, ръководител на групата по приложна математика в университета на Манчестър Дата: 11.09.2024 г. Час: 14:00 ч. Място: зала 403 Резюме: An interesting feature of subdiffusion-reaction systems is the lack of a universal model for reactions in subdiffusive media. Consequently, the coarse-grained fractional subdiffusion-reaction equations depend on the specific details of the underlying microscopic random walk models. This complexity arises from the memory effects inherent in subdiffusive transport systems. Simply adding a reaction term to the transport equation can be physically inconsistent. Due to the memory effect, the subdiffusion process depends on the entire history of the system’s evolution making the diffusion and reaction processes inseparable.  

An efficient approach for searching three-body periodic orbits passing through Euler configuration Докладчик: Иван Христов (ФМИ-СУ) Дата: 28.05.2024 г. Час: 15:00 ч. Място: зала 503 Резюме: The classic gravitational three-body problem is considered. The motion of the bodies is determined by Newton’s second law and Newton’s law of universal gravitation. We are interested in a special case of planar periodic orbits – those that pass through Euler configuration. In this talk a new efficient approach for searching three-body periodic equal-mass collisionless orbits passing through Euler configuration is presented [1]. The approach is based on a symmetry property of the solutions at the half period. Depending on two previously established symmetry types on the shape sphere, each solution is presented by one or two distinct initial [...]

On constructions of spherical codes Докладчик: д-р Константин Воробьов, ИМИ - БАН Дата: 27.03.2024 г. Час: 16:00 ч. Място: зала 503 Резюме: In this talk, we review various approaches to construct spherical codes with a given coding distance. We also propose a new approach that allows us to find new codes on spheres in small dimensions. In particular, we find a previously unknown kissing arrangement of 40 spheres in R^5.  

Temporal networks, Tensors, and applications to ML Докладчик: проф. Огнян Кунчев, ИМИ - БАН Дата: 31.01.2024 г. Час: 15:00 ч. Място: ... Резюме: Recently, temporal networks got an increasing interest, related to their numerous applications in real-life data analysis. The temporal networks (more mathematically, “graphs depending on time”) are a typical example of a Tensor – the adjacency matrix depends on several parameters and is a tensor (in the mathematical sense of the word). We present some recent results about the structure of the spectrum of the normalized Laplacian in the case of “slowly varying temporal networks”. The detailed structure of the spectrum gives rise of the notion of “constant block Jacobi model”, which has closed-form solution for the spectrum and the eigenfunctions. This [...]

How log-concavity fought lottery frauds in Florida and what it did after winning Докладчик: Дмитри Карп Дата: 14.11.2023 г. Час: 14:00 ч. Място: Зала 503 на ИМИ-БАН Резюме: In the talk I will first discuss the log-concavity in parameters of the normalized beta function (which represents the cumulative distribution function of beta distributed random variable) that has been used recently in an investigation of lottery frauds in Florida.  I will then present some related  properties of incomplete beta function and proceed to general log-concavity problems and their particular cases motivated by various problems in statistics, probability and analysis. Фейсбук страница на семинара: Семинар по Приложна математика

Крайни автомати и невронни мрежи за езиково моделиране Докладчик: проф. Стоян Михов Дата: 25.10.2023 г. Час: 15:00 ч. Място: Зала 503 на ИМИ-БАН Резюме: Езиковите модели са основен инструмент за решаването на редица сложни задачи като извличане на информация, генерация на естествен език, машинен превод, речева комуникация и много други. Традиционно езиковите модели се представят с крайни преобразуватели. През последните години обаче езиковите модели, базирани на невронни мрежи, извършиха значителен скок в качеството на езиковото моделиране, с което се постигнаха близки до човешки способности в много задачи за разбиране на естествен език. В предстоящата лекция ще направим първо кратък преглед на традиционните представяния на езикови модели, базирани на крайни преобразуватели. След това ще покажем често използвани архитектури на невронни мрежи за езиково моделиране и ще [...]