Abstract
Complexity of communication networks is ever increasing and getting complicated by their heterogeneity and dynamism. Traditional techniques are facing challenges in network performance management. Cognitive networking is an emerging paradigm to make networks more intelligent, thereby overcoming traditional limitations and potentially achieving better performance. The vision is that, networks should be able to monitor themselves, reason upon changes in self and environment, act towards the achievement of specific goals and learn from experience.
The concept of a Cognitive Engine (CE) supporting cognitive functions, as part of network elements, enabling above said autonomic capabilities is gathering attention. Awareness of the self and the world is an important aspect of the cognitive engine to be autonomic. This is achieved through embedding their models in the engine, but the complexity and achievable truthfulness of such models are of concern considering the dynamic and non-linear behavior of the network. Moreover the knowledge model should be able to capture and represent the holistic aspect of the network in a scalable manner.
In the present work, I focus on the architectural aspects of the cognitive engine that incorporates a context space based information structure to its knowledge model. I propose a set of guiding principles behind a cognitive system to be autonomic and use them with additional requirements to build a detailed architecture for the cognitive engine. I define a context space structure integrating various information structures that are required for the knowledge model. Use graphical models towards representing and reasoning about context space is a direction followed here. Specifically I analyze the framework of qualitative models for their suitability to represent the dynamic behavior of the wireless network. The motivation behind this novel approach is in the possibility of building the knowledge model from the qualitative information in the form of influence diagrams elicited from human experts. Considering the difficulties of building large scale models through structure learning, the above approach is attractive. After a detailed analysis of the qualitative model I select a set of fitting semi-qualitative extensions with inference mechanisms to overcome their observed limitations. With learning from this exercise I propose a methodology for preparing and using the qualitative models in a cognitive engine.
Further I use the methodology in multiple functional scenarios of cognitive networks including self- optimization and self- monitoring. In the case of self-optimization, I integrate principles from monotonicity analysis to evaluate and enhance qualitative models as part of the methodology. Related to self-monitoring, the proposal is on an architecture for network monitoring and fault diagnostics using qualitative models.
Towards the end I propose a novel cognitive acoustic communication network for short range data communication between devices. I present the design and implementation details along with its interesting applications for near field communications. Further I compare the qualitative models of the acoustic network with an equivalent radio network. The comparison results points to the generality of qualitative models across multi-technology systems.
Major contributions of this research work is in discovering the applicability of qualitative knowledge models and developing mechanisms to efficiently use them in cognitive networks.
The concept of a Cognitive Engine (CE) supporting cognitive functions, as part of network elements, enabling above said autonomic capabilities is gathering attention. Awareness of the self and the world is an important aspect of the cognitive engine to be autonomic. This is achieved through embedding their models in the engine, but the complexity and achievable truthfulness of such models are of concern considering the dynamic and non-linear behavior of the network. Moreover the knowledge model should be able to capture and represent the holistic aspect of the network in a scalable manner.
In the present work, I focus on the architectural aspects of the cognitive engine that incorporates a context space based information structure to its knowledge model. I propose a set of guiding principles behind a cognitive system to be autonomic and use them with additional requirements to build a detailed architecture for the cognitive engine. I define a context space structure integrating various information structures that are required for the knowledge model. Use graphical models towards representing and reasoning about context space is a direction followed here. Specifically I analyze the framework of qualitative models for their suitability to represent the dynamic behavior of the wireless network. The motivation behind this novel approach is in the possibility of building the knowledge model from the qualitative information in the form of influence diagrams elicited from human experts. Considering the difficulties of building large scale models through structure learning, the above approach is attractive. After a detailed analysis of the qualitative model I select a set of fitting semi-qualitative extensions with inference mechanisms to overcome their observed limitations. With learning from this exercise I propose a methodology for preparing and using the qualitative models in a cognitive engine.
Further I use the methodology in multiple functional scenarios of cognitive networks including self- optimization and self- monitoring. In the case of self-optimization, I integrate principles from monotonicity analysis to evaluate and enhance qualitative models as part of the methodology. Related to self-monitoring, the proposal is on an architecture for network monitoring and fault diagnostics using qualitative models.
Towards the end I propose a novel cognitive acoustic communication network for short range data communication between devices. I present the design and implementation details along with its interesting applications for near field communications. Further I compare the qualitative models of the acoustic network with an equivalent radio network. The comparison results points to the generality of qualitative models across multi-technology systems.
Major contributions of this research work is in discovering the applicability of qualitative knowledge models and developing mechanisms to efficiently use them in cognitive networks.
| Original language | English |
|---|---|
| Publication status | Published - 2013 |