Poster Abstract: Context Intelligence in Pervasive Environments

Intelligent personalization systems are becoming increasingly reliant on contextually-relevant devices and services, such as those available within modern IoT deployments. An IoT context may emerge-or become pervasive-when the intelligent system generates knowledge from dialogue-based interactions with the end-user; the context is strengthened even further by incorporating state representations about the environment (e.g., generated from wireless sensor data) into the knowledge graph. This is crucial for pervasive applications like digital assistance in IoT, where context-aware systems need to adapt quickly: activities like leaving work home-bound, driving to the grocery store, arriving at home, and walking the dog, for example, can occur in a relatively short period of time-during which an intelligent assistant must be able to support user requests in a consistent and coherent manner. Given that computational ontologies can serve as semantic models for heterogeneous data, they are becoming increasingly viable for reasoning across different IoT contexts. This involves: (a) federation and dynamic pruning of multiple modular ontologies, ideally, to comprehensively capture only the knowledge that will facilitate execution of a multi-context task; (b) fast consistency-checking and ontology-based inferences, aided by rules-based execution environments that can evaluate/transform ambient wireless sensor network (WSN) data, in real-time; and (c) run-time execution of ontology-based control procedures, through rule-engine actuation commands sent across the WSN. Only by realizing these functionalities may intelligent systems be capable of reasoning over device properties, system states, and user activities, while appropriately delegating commands to other intelligent agents or other relevant IoT services. In this poster, we illustrate how a multi-context knowledge base can be structured on the basis of modular ontologies and integrated with a distributed rules-based inference engine in multiple smart-building environments, in order to enable scalable contextual reasoning for intelligent assistance. Preliminary results are also discussed. This work is conducted through the partnership of Bosch Research Pittsburgh and Carnegie Mellon University (CMU), and is in partial satisfaction of CMU's Bosch Energy Research Network (BERN) grant, awarded for developments in intelligent building solutions. The approach we describe is also partially based on the Ubiquitous Personal Assistant (UPA) project, Bosch Research's largest research initiative worldwide.