Seminar LAMBDA: Online Neural Networks for Change-Point Detection

Аннотация: 

Change points refer to moments when a time series shifts its underlying behavior. Detecting such a change point indicates that the system's state has changed, and early detection can help avert negative outcomes. In this talk, we will present our recently published paper, which introduces two change-point detection methods based on neural networks and online learning.

Both methods have linear computational complexity, making them well-suited for detecting change points in large-scale time series. We evaluate their performance against state-of-the-art algorithms using a range of synthetic and real-world datasets. Additionally, we provide theoretical proofs of convergence to optimal solutions and outline the conditions under which our approach outperforms offline detection methods.

Место проведения: АУК Покровский бульвар, 11, ауд. R408
Дата: 20.04.2026 
Время: 14:40-16:00