Aimilia Bantouna

An overview of learning mechanisms for cognitive systems

Cognitive systems were first introduced by Mitola and in the last decade they have proved to be beneficial in self-management functionalities of future generation networks. The advantages and the way that networks gain benefits from cognitive systems is analysed in this article. Moreover, since such systems are closely related to machine learning, the focus of this article is also placed on machine learning techniques applied both in the network and the user devices side. In particular, celebrating 10 years of cognitive systems, this survey-oriented article presents an extended state-of-the-art of machine learning applied to cognitive systems as coming from the recent research and an overview of three different learning capabilities of both the network and the user device.

Operator-driven framework for establishing and unifying autonomic network and service management solutions

It is commonly recognized that the technology progress, dynamism but also complexity of telecommunication networks and services increase with rapid paces. Such challenges cannot be efficiently handled by traditional networking and management schemes. Autonomics in network and services management appear as the most viable way out. However, despite the significant research efforts and achievements in this field, a few and only recently start to convince operators for their deployability. In this direction, UniverSelf is a research initiative which proposes a pragmatic solution for overcoming the increasing complexity a) by consolidating and capitalizing on lessons learnt and b) by identifying and solving actual, first-priority, immediate and mid-term manageability problems encountered by operators. The cornerstone of UniverSelf approach is the Unified Management Framework (UMF), an operator-driven framework that designates processes, tools and methods for establishing (legacy, emerging and yet undiscovered) autonomic solutions in the joint management of networks and services. This paper provides a first concise description of the UMF design in terms of core, reusable and cohesive functional blocks and interfaces, as derives from the elaboration of requirements elicited from a set of operator problems (use cases). The design is complemented by principles and goals that address important high-level challenges such as the unification/federation of diverse autonomic solutions and technology domains, the governance of autonomic infrastructures and services, as well as the embodiment of autonomic solutions (intelligence) into the management ecosystem.

Self-Organizing Maps for advanced learning in cognitive radio systems

Modern everyday life keeps making wireless communications more and more popular. The wireless communications landscape is highly varying and its success depends on the efficient provision of a physically limited natural source namely, radio spectrum. Cognitive radio systems (CRSs) have been proposed as a very promising technology for addressing this situation by facilitating more flexible and intelligent spectrum management. However, the processes of a CRS are often proved to be rather arduous and time consuming. Accordingly, a learning mechanism, capable of building knowledge to the system can speed up the whole cognition process. Framed within this statement, this paper introduces and evaluates a mechanism which is based on the well-known unsupervised learning technique, called Self-Organizing Maps (SOMs), and is used for assisting a CRS to predict the raw data rate that can be obtained, when it senses specific input data from its environment. Results show that the proposed method can provide predictions which are correct up to a percentage of 78.9% while exhibiting performance comparable to other supervised neural network-based learning schemes.

A survey of autonomic networking architectures: towards a Unified Management Framework

SUMMARY Academic and industrial research initiatives have sought to make fully autonomic networks a reality. Some of these initiatives pursued a holistic approach, while others focused on setting up functionalities for specific networking domains. These efforts did not succeed in being extensively deployed, because the goals of network operators were not satisfactorily met. These goals include unification of management operations, enablement of end-to-end management and enhancement of the overall system performance in a trusted way, while reducing management cost. In this paper, we analyse a set of existing autonomic management architectures and frameworks with respect to a selected set of criteria. We then identify missing parts and challenges and propose a framework to unify the most promising attributes towards a novel approach of realization of autonomic networking management. We call this proposal Unified Management Framework (UMF). Copyright

Knowledge Management Toolbox: Machine Learning for Cognitive Radio Networks

Learning mechanisms are essential for the attainment of experience and knowledge in cognitive radio (CR) systems, exposed to high dynamics with often unpredictable states [1]. These mechanisms can be associated with user and device profiles, context, and decisions. The focus learning user preferences is the dynamic inference and estimation of current and future user preferences. The acquisition and learning of context information encompasses mechanisms for the system to perceive its current status and conditions in its present environment, as well as estimating (and forecasting) the capabilities of available network configurations. Finally, learning related to decisions addresses the building of knowledge with respect to the efficiency of solutions that can be applied to specific situations encountered. Based on knowledge obtained through learning, decision-making mechanisms can become faster, since the CR system can learn and immediately apply solutions that have been identified as being efficient in the past. Moreover, knowledge obtained through learning mechanisms may be shared among nodes of a system. Thus, more reliable and more optimal decisions can be made by exploiting knowledge obtained through learning mechanisms.

Network Load Predictions Based on Big Data and the Utilization of Self-Organizing Maps

The pervasiveness of computers in everyday life has already increased and keeps increasing the available digital data both in volume and variety/disparity. This large and dynamic availability of digital data is referred to as Big Data and is very promising in bringing forward new insights and knowledge. For obtaining these insights, the proper combination and processing of the data is required. However, the dynamicity and the increasing size of data start making their handling impossible for analysts and raise many concerns on the manner in which data will be processed from now on. Towards this direction, this paper proposes a tool that processes and combines disparate data in order to create insights regarding a future network load. In particular, the tool (based on the unsupervised machine learning technique of Self-Organizing Maps) builds knowledge on the network load that is encountered with respect to the date of interest, the location, the weather, and the features of the day (e.g., weekend, bank holiday, etc.). The obtained results reveal that the tool is capable of learning the traffic pattern and thus predicting the network load that will be encountered in the near or distant future given information for the above presented parameters with small deviations (up to 0.000553 in terms of Mean Square Error). Moreover, the tool maintains only the most representative data instances and thus reduces the data storage requirements with no loss of information.

Cross-Layer Analysis in Cognitive Radio—Context Identification and Decision Making Aspects

Research on context-aware communications has recently led to the introduction of features and algorithms relying on the presence of rich, accurate context information, and requiring however, the introduction of cross-layer information exchanges. Cognitive radio (CR), in particular, is expected to benefit from context awareness, as the cognitive engine (CE) relies on the availability of multiple information sources to operate efficiently. In this context, this work delivers a detailed, yet concise classification and description of the information exchanged in a CR network between the layers of a generic protocol stack, and between each layer and the CE. For each layer, the key services provided and delivered are presented, followed by a catalogue of exchanged parameters. The analysis, supported by a set of use cases providing a quantitative assessment of the impact of crosslayer information exchanges in a CR framework, is the basis for the discussion of key implementation challenges and the identification of the most promising partition of functions and tasks between layers and CE.

Role of neighbour discovery in distributed learning and knowledge sharing algorithms for cognitive wireless networks

This work investigates the impact of neighbour discovery on distributed learning schemes applied on optimal network selection based on the acquisition by the selecting device of context information on the capabilities and status of surrounding networks. The work introduces the problem of neighbour discovery in multiple channel and cognitive networks, and identifies the trade-offs between neighbour discovery performance and overall network performance. Next, an optimal network selection algorithm based on distributed learning is introduced, and key parameters and components relevant to its operation are presented, focusing in particular on the common control channel required to exchange the context information. Finally, the paper discusses the relation between neighbour discovery and the distributed learning process at the basis of the context information acquisition; a model for mapping the learning process on a neighbour discovery problem is proposed, and the potential impact of neighbour discovery failures on the performance of the optimal network selection scheme is discussed.

Unifying management of future networks with trust

Stability, robustness, and security issues arising from future self-organizing networks (SONs) must be understood today, in order to be incorporated into their design, standardization, and certification. We address the issue of operator trust in Long Term Evolution (LTE) SON through the following five requirements and outline our approaches to meet them: 1) trust must be measurable, 2) trust must be SON-specific, 3) trust must be model-driven, 4) trust must be propagated end-to-end, and 5) trust must be certified. As such, we consider the three facets of operator trust — reliable operation, trustworthy interworking, and seamless deployment and suggest a composite metric for SON stability; we define a key performance indicator (KPI)-based envelope of dependable adaptations; we demonstrate how to construct such models based on predicates; we show that trust networks emerge from predicate-enabled behaviors; and we outline the certification process. Trust predicates that are defined at the design phase as abstract behaviors, and verified at runtime as fully qualified ones, prove to have the power of policies. Once checked, they can be reused many times, and rewritten to cater to new behaviors.

Mechanisms for information and knowledge sharing in wireless communication systems

Information and knowledge sharing mechanisms are essential for the efficient operation of cognitive wireless communication systems, complementing learning and decision making and enabling the coordination between various management entities. Information and knowledge sharing mechanisms can be defined as means for transmitting elements of information necessary to manage and realize various operations within future cognitive wireless communications systems. Information may be conveyed from network elements to terminals and vice versa, and may also be exploited for the exchange of information between terminals, so as to increase the accuracy of obtained knowledge. The paper provides an overview of related work and describes several cases of how information and knowledge sharing mechanisms can be exploited in wireless communication systems. A detailed description of the information flow between various entities is provided. An overview of potential implementation options is provided.