TAIRCOMP

TAIRCOMP - Turning the Air into an AI Computer
Wireless networks are becoming overwhelmed by the huge data traffic that stems from the emerging artificial intelligence (AI) and machine learning (ML) applications. The complex AI/ML computations will generate traffic flows that cannot be supported by the current network designs. A fundamental re-design of these networks is needed. To address such issue, over-the-air (OtA) computation is a ground-breaking concept that exploits the superposition principle of wave propagation governed by Maxwell’s equations. Thereby, the wireless channel can execute the AI/ML computations along with the communications. How to perform AI/ML by OtA is a largely open research question which we will address in this research project. The establishment of OtA methods for AI/ML can have a profound significance in science and technology. Within science, OtA will require going beyond classical Shannon information theory, which has underpinned all generations of networks so far. Our research project will establish a new theoretical and algorithmic framework for wireless communications for computations. Within technology, if fundamentally new design methods will not be established, future massive demand of AI/ML will crush the current networks, because the AI/ML traffic will be energy unsustainable, will be supported only by few devices, and will quicky saturate the available spectrum. To reach the ambitious goals of the project, we have planned three main work packages centred around the foundations of wireless communications for OtA, their use for the development of AI/ML methods and algorithms by OtA, and the investigation of these algorithms’ security and robustness. The team is composed of researchers with interdisciplinary competences in wireless communications, optimization, and AI/ML. The project will be performed in close collaboration with major industrial and academic research centres such as Ericsson Research, Princeton University, MIT and University of California at Berkeley.