Семинар LAMBDA: Generative flow networks for modeling highly structured objects

Александра Волохова (Mila & Université de Montréal) выступит с докладом "Generative flow networks for modeling highly structured objects" на еженедельном семинаре LAMBDA.

Семинар будет проходить 10 июня в 12:00

Аннотация

Generative flow network (GFlowNet) is a machine learning model designed for generating highly structured objects like molecules, graphs, etc. After being trained, this model can generate a diverse set of samples from the target distribution, which can be defined by an energy function. This diversity is particularly important when the energy function has multiple local minima, which often poses a challenge for other sampling techniques, such as Gibbs sampling. In this talk, I will show you how GFlowNet works and how it can be trained with either known or unknown target energy function.
 

Время: 12:00-13:00 (UTC+3)
Место:  Онлайн, платформа Zoom
Язык: английский
ZOOM: