The ambition of the ELEPHY project is to boost the development of a nascent research field at the intersection of AI and machine learning-aided communications engineering. While the traditional approach to communication systems design has been based on human-engineered schemes to reliably convey bit messages from one location to another over noisy or rate-limited links, some recent research directions have expanded this scope to a multi-party remote interaction game where the rules, semantics, costs and rewards are not fixed a priori, but negotiated and emerged on the fly.
The ELEPHY project will build upon recently proposed deep-learning (DL) and reinforcement learning (RL) concepts for end-to-end (E2E) learning of physical-layer (PHY) communication systems. The ambition of ELEPHY is to expand the scope of E2E learning to encompass the emergence of some basic medium access control (MAC) or higher-layer functionalities (e.g., retransmissions, addressing and identification, authentication, scheduling, routing, link establishment, CSI reporting, feedback schemes) by leveraging cooperative multi-agent learning.
ELEPHY aims to contribute towards a longterm vision which consists of a multi-party communication system in a dynamic environment (e.g., in the context of IoT, self-driving vehicles, sensor networks, or similar) in which the participating devices cooperatively learn an efficient machine language which emulates (and surpasses) the multiple functions of the L1/L2/L3 layer stack that have been traditionally designed by hand.
We expect the successful candidate to have solid knowledge of communication systems and AI/ML techniques, and be capable to pick up quickly on new tools, AI architectures and cutting-edge techniques.
The activity will mostly involve empirical research via simulations as well as theoretical study. The ISPIC lab has powerful computational resources, including multiple GPU-equipped workstations for simulations. Additional hardware experiments at the ISPIC labs (e.g., the development of an over-the-air proof-of-concept that demonstrates protocol emergence with multiple USRPs) may be considered too, depending on the profile of the candidate.
Closes: 10/09/2024
Més posts de Recerca
Comentaris