Заседание научного семинара лаборатории ЛАТАС

Докладчик: Nikita Severin (HSE University and Institute for System Programming  RAS)
Тема: From Static to Dynamic Graphs: Towards Edge-Centric Learning
Язык семинара: English

Graphs are at the core of many real-world systems, from social networks to financial transactions. This talk is based on PhD thesis of the speaker and explores machine learning methods on graphs in both static and dynamic settings. The first part highlights hidden risks in widely used static graph neural networks, showing how sensitive information can be unintentionally revealed. Then, the focus turns to dynamic graphs, which better capture the changing nature of real systems. The talk will introduce a unified benchmarking framework for evaluating dynamic models is introduced, providing a clearer picture of their true capabilities. Finally, new approaches are presented that place edges, rather than nodes, at the center of learning, enabling models to capture fine-grained temporal patterns often overlooked by existing methods.

Приглашаются все желающие!

Явка аспирантов школы компьютерных наук обязательна.