A general cognitive framework for context-aware systems: extensions and applications for high level information fusion approaches

• Autores: Miguel A. Serrano
• Directores de la Tesis: Miguel Ángel Patricio Guisado (codir. tes.) , Jesus Garcia Herrero (codir. tes.)
• Lectura: En la Universidad Carlos III de Madrid ( España ) en 2016
• Idioma: inglés
• Tribunal Calificador de la Tesis: José Manuel Molina López (presid.) , Ángel Arroyo Castillo (secret.) , Nayat Sánchez-Pi (voc.)
• Enlaces
  • Tesis en acceso abierto en: e-Archivo
• Resumen

  • Context-aware systems aims at the development of computational systems that process data acquired from different datasources and adapt their behaviour in order to provide the 'right' information, at the 'right' time, in the 'right' place, in the 'right' way to the 'right' person (Fischer, 2012). Traditionally computational research has tried to answer these needs by means of low-level algorithms. In the last years the combination of numeric and symbolic approaches has offered the opportunity to create systems to deal with these issues. However, although the performance of algorithms and the quality of the data directly provided by computers and devices has quickly improved, symbolic models used to represent the resulting knowledge have not yet been adapted to smart environments. This lack of representation does not allow to take advantage of the semantic quality of the information provided by new sensors. This dissertation proposes a set of extensions and applications focused on a cognitive framework for the implementation of context-aware systems based on a general model inspired by the Information Fusion paradigm. This model is stepped in several abstraction levels from low-level raw data to high level scene interpretation whose structure is determined by a set of ontologies. Each ontology level provides a skeleton that includes general concepts and relations to describe entities and their connections. This structure has been designed to promote extensibility and modularity, and might be refined to apply this model in specific domains. This framework combines a priori context knowledge represented with ontologies with real data coming from sensors to support logic-based high-level interpretation of the current situation and to automatically generate feedback recommendations to adjust data acquisition procedures. This work advocates for the introduction of general purpose cognitive layers in order to obtain a closer representation to the human cognition, generate additional knowledge and improve the high-level interpretation. Extensibility and adaptability of the basic ontology levels is demonstrated with the introduction of these traverse semantic layers which are able to be present and represent information at several granularity levels of knowledge using a common formalism. Context-based system must be able to reason about uncertainty. However the reasoning associated to ontologies has been limited to classical description logic mechanisms. This research also tackle the problem of reasoning under uncertainty circumstances through a logic-based paradigm for abductive reasoning: the Belief-Argumentation System. The main contribution of this dissertation is the adaptation of the general architecture and the theoretical proposals to several context-aware application areas such as Ambient Intelligence, Social Signal Processing and surveillance systems. The implementation of prototypes and examples for these areas are explained along this dissertation to progressively illustrate the improvements and extensions in the framework. To initially depict the general model, its components and the basic reasoning mechanisms a video-based Ambient Intelligence application is presented. The advantages and features of the framework extensions through traverse cognitive layers are demonstrated in a Social Signal Processing case for the elaboration of automatic market researches. Finally, the functioning of the system under uncertainty circumstances is illustrated with several examples to support decision makers in the detection of potential threats in common harbor scenarios.