Description
Date depot: 17 septembre 2019
Titre: C-Continuum: Edge-to-Cloud computing for distributed AI
Directeur de thèse:
Giovanni PAU (BSS)
Encadrant :
Luigi LUIGI ATZORI (Univ. di Cagliari)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
Mobile autonomous systems are supposed to deeply impact in manufacturing, space exploration, rescue, defense, transportation, and everyday life. Autonomous air-ground vehicles, for example, will become normal tools in the next few years, providing a natural platform for distributed artificial intelligence applications including, for example, disaster rescue and recovery, area surveying, autonomous driving, etc. The raise of autonomous cooperating robots will pose new challenges in networking, distributed systems and resource management. Heavy computational tasks will be dispatched to the closest edge node for processing and the core-cloud will be involved as last resort in an effort to reduce latency and increase the global system capacity leveraging application and resource locality. Massive amounts of data and computations will be required. For example, in the autonomous driving scenario Intel estimates that each driver-less vehicle will produce over 4 TeraBytes of data each day. While most of this data is consumed in-car, cooperating autonomous vehicles will have to exchange some percentage of the 4TB and eventually off-load some computation and data to the local edge or the core-cloud. This is particularly relevant when locally gathered and labeled data can be used to refine the model and ultimately increase the global 'intelligence'. This approach is often taken by autonomous driving automakers.
Distributed AI applications call for effective, seamless, and efficient communication and computation mechanisms across the whole computing spectrum edge, fog, and cloud. Computational tasks, AI models and relative data may be instantiated on ground vehicles, in air-drones, and/or in the local edge DC. The allocation may and will change frequently during a single execution.
Here, I outrageously propose C-Continuum, a Computing Continuum framework targeting distributed AI in the mobile arena. C-Continuum aims to define a new generation of tools and mechanisms designed to enable fine-granularity computation, coordination and mobility management across the mobile-computing spectrum from the edge to the core. C-Continuum embraces Named Data Networking (NDN) making a case for naming any computational entity and using those names for resource location, data transfers, and computing function
Doctorant.e: Aguiari Davide