У оквиру традиционалног колоквијума Института за физику у Београду, у четвртак, 2. новембра 2023. године, у 12 часова у сали „Звонко Марић“ предавање под насловом:
Emulating Recurrent Neural Network Processors via Nonlinear Delay Dynamics
одржаће проф. др Лоран Ларже (FEMTO-ST, истраживачки институт Француског националног центра за научна истраживања, регион Бургундија-Франш-Конте, Француска).
САЖЕТАК:
Delay dynamical systems are poorly popular in science, however being present in many situations of the real world and real life. They are intrinsically infinite dimensional system, and they are even capable of generating high complexity chaotic solutions when non-linearities are involved. After a general introduction of nonlinear delay dynamics, examples of real life, characteristic theoretical features, and some practical applications in photonics they were already used for (optical chaos cryptography, high spectral purity microwave source for radar), the talk will introduce the way nonlinear delay dynamics can actually also emulate a network of neurons, which can further be used for designing neuromorphic processors according to the Reservoir Computing concept. Motivations and experimental demonstrations will be presented in the framework of ultra-fast and energy efficient hardware dedicated to Artificial Intelligence processors.