logo EDITE Sujets de doctorat

Autonomic Mechanisms for IoT services

Résumé rédigé par
Directeur de thèse:
Doctorant: Nesrine AMMAR
Unité de recherche UMR 7606 Laboratoire d'informatique de Paris 6


In order to make this vision a reality, one of our current research objectives is the design of some autonomic functions that could be part of a digital assistant (software) to help end-users and smart space operators in identifying, setting-up and improving new IoT services that use various connected objects. This goes beyond the connectivity aspect: How to characterize IoT services? How to help people in the choice of the connected devices for their service? How to make the digital assistant self-recommend IoT services to the end-users? The PhD thesis will contribute to our research project by exploring and proposing new autonomous mechanisms within the three following research axes: 1. Self-modeling of IoT services. The digital assistant manipulates Virtual Objects and Chains of Virtual Objects that are respectively the abstractions of the connected objects in the real world and the abstractions of the way they interact together in IoT services. We already investigated typed attributed graphs [9], but this modeling is static and requires prior knowledge. We would like to go towards self-modeling of IoT services (a bit like what is done in robotics, see for example [10]), incorporating time aspects (e.g., using automata [11]) and contextualized information (e.g., by monitoring trac ows between connected objects, see for example [12]). 2. Self-characterization of IoT services. The digital assistant should be able to autonomously \understand" what the IoT services in its scope are and how they are working, self-building an IoT service catalog. Based on the static modeling of axis 1, we already de ned a rst algorithm that gives a signature for each service instance [patent led and publication under submission]. How to adapt/improve/modify it with dynamic (i.e., including time and spatial dimensions) and contextualized information? Do we need to use clustering [13] or machine learning techniques [14]? 3. Self-adaptive mechanisms for IoT services. The IoT service catalog of axis 2 constitutes the knowledge of the digital assistant, which can be used to help people in the usage of new IoT-based services and IoT service administrators in the deployment of IoT services. A bit like in Autonomic Computing [15], one can apply Arti cial Intelligence techniques [16] on this knowledge to analyze events and take some decisions, learning from past decisions [14]. Autonomic mechanisms will be then de ned using models of axis 1 and self-characterization mechanisms of axis 2. Here are some examples of such mechanisms: (a) Self-healing { Based on learning on existing and working IoT service instances, the digital assistant should autonomously detect any problem in some new instance of an IoT service that some user(s) would like to set-up, diagnose the problem (e.g., missing or wrong Virtual Object in the Chain of Virtual Objects of the service), and repair (e.g., proposing to add a new Virtual Object to the Chain of Virtual Objects to make the service work). (b) Self-improving mechanisms { Some connected devices may be missing when several IoT services are requested. The system should be able to propose some alternative solutions taking into account the requests while minimizing the induced cost. Some resources may be preempted in unused or low-priority IoT services to complete higher priority new requested services (e.g., e-health scenarios, emergency scenarios). (c) Gracefully degrading mechanisms { When some connected devices are failing, some IoT services can be a ected. Degraded service behavior may be envisaged when some services are critical (e.g., home surveillance), notifying end-users that the full service is not available but only part of it. (d) Other { Some other possible autonomic mechanisms can be envisaged, e.g., self-coordination of IoT services.