Daniel Lathrop: Using nonlinear dynamics for low-power high-speed machine learning electronics
Автор: Machine Learning and Dynamical Systems Seminar
Загружено: 2026-02-23
Просмотров: 26
Описание:
Title: Using nonlinear dynamics for low-power high-speed machine learning electronics
Speaker: Daniel Lathrop (University of Maryland)
Abstract:
Dynamical systems are a rich playground for the invention of beyond von Neumann computational systems. I'll discuss two approaches here: using logic gates as reservoir computers for inference and using stochastic p-bits as Ising computers. The first system utilizes the intrinsic nonlinearity of CMOS logic gates as compute elements in large reservoir graphs. We have successfully tested these designs in classification problems as well as documented their low power usage. The second example uses interesting stochastic dynamics of magnetic tunnel junctions as probabilistic bits (p-bits). We have extensively tested these devices toward using them in novel designs for Ising computers to solve NP-hard problems.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
