Deployment of Intelligent Agents in Cognitive networks
IJCSIS is an open access publishing venue for research in general computer science and information security. Target Audience: IT academics, university IT faculties; industry IT departments; government departments; the mobile industry and computing industry. Coverage includes: security infrastructures, network security: Internet security, content protection, cryptography, steganography and formal methods in information security; computer science, computer applications, multimedia systems, software, information systems, intelligent systems, web services, data mining, wireless communication, networking and technologies, innovation technology and management. The average paper acceptance rate for IJCSIS issues is kept at 25-30% with an aim to provide selective research work of quality in the areas of computer science and engineering. Thanks for your contributions in September 2010 issue and we are grateful to the experienced team of reviewers for providing valuable comments.
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 Deployment of Intelligent Agents in Cognitive Networks Huda Fatima Dr.Sateesh Kumar Pradhan Mohiuddin Ali Khan Dr. G.N.Dash Dept. of CS Dept. of Comp.Engineering Dept. of Comp. Networks Dept. of Comp. Science Jazan University King Khalid University Jazan University Sambalpur University Jazan, K.S.A Abha, K.S.A Jazan, K.S.A Orissa, India email@example.com firstname.lastname@example.org email@example.com firstname.lastname@example.org Abstract—Every organization faces challenging maintainable and their capabilities should be task in the designing of the communication continuously improved and upgraded by relying as network in order to make its efficiency smoother little as possible on human intervention. Therefore by the increasing complexities. Therefore, we the network research community proposed a new have to proposed a concept of cognitive network and how the intelligent agents are deployed to concept of networking: The Cognitive Network. overcome the challenges.With the tremendous What is a Cognitive Network and how are the expansion of networks across the globe, the intelligent agents deployed is what we have deployment of intelligent agents in cognitive presented here. networks contributes as an efficient, reliable and challenging task for the researchers. In this Cognitive networks paper, we survey the existing research work on cognitive networks and later we provide the In this section, we analyze several existing artificial intelligent techniques that are definitions for cognitive networks, and we argue that potentially suitable for the development of two elements are essential for developing a cognitive cognitive networks. network (CN): the knowledge representation and the Keywords: Artificial Intelligence, Cognitive cognition loop. Next, we discuss the framework network, Intelligent agents. proposed in  for introducing cognition to communication networks. The main part of the section focuses on methods from AI that seem applicable for developing CNs. We provide a I. INTRODUCTION: summary of several types of intelligent agents (IAs), map them to the functional states of the cognitive One of the fastest growing areas is the information loop. As we go along, we also refer to existing and communication technologies. These changes research on CNs which makes use of the respective have an immediate impact on diverse aspects of the type of IA, where available. How it started? The modern society, which includes inter-human word cognitive refers to an entity that is able to relations, economy, education & entertainment. In perform some kind of conscious intellectual activity this respect, the development of reliable, flexible and such as thinking, reasoning, learning or remembering future-proof infrastructure should be capable of in order to make sense of its surroundings. This word increasing the users’ quality of life by providing was first used in communication networks to refer to services such as e-health, e-learning and e-payments. a technology by Mitola as he introduced the In order to meet the demand of the increased cognitive radio . complexity, future networks should be easily 122 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 distributed cognitive entities (agents) which are somehow “smart” as they have certain reasoning We would like to emphasize that, according to the capabilities to be connected to the network. The dictionary , the word cognitive used as an entities in such a network interact with each other, adjective to a noun means: of, relating to, being, or they can cooperate, act selfishly or a combination of involving conscious intellectual activity (as thinking, the two. While functioning in this environment, the reasoning, or remembering); based on or capable of entities are able to learn and take decisions in such being reduced to empirical factual knowledge. way as to reach an end-to-end goal. These end-to- end goals are dictated by the business and user In , the authors define the CN as a network with a requirements [2,4]. Developing and maintaining such cognitive process that can perceive current network a network is an extremely challenging task and has conditions, plan, decide, act on those conditions, enormous potential, especially in the area of network learn from the consequences of its actions, all while management. following end-to-end goals. This loop, the cognition loop, senses the environment, plans actions A Cognitive Network needs to evolve overtime: its according to input from sensors and network set of technologies has to be updated by removing policies, decides which scenario fits best its end-to- deprecated and adding new ones; its set of tools that end purpose using a reasoning engine, and finally help managing complexity should be added and acts on the chosen scenario as discussed in the removed in a plug and play fashion. Thus, the previous section. The system learns from the past architecture of cognitive network should be flexible (situations, plans, decisions, actions) and uses this and should lead to a modular and highly scalable knowledge to improve the decisions in the future. infrastructure. Furthermore, the cognitive network must be self aware : it should be able to determine This definition of CN does not explicitly mention the appropriate actions to achieve goals and to learn knowledge of the network; it only describes the while doing all these. It should be self-configuring, cognitive loop and adds end-to-end goals that would self-optimizing, self-healing and self-protecting in a distinguish it from CR or so called cognitive layers cognitive way. . We consider this definition of CN incomplete since it lacks knowledge which is an important In this paper, we analyze some recent trends in the component of a cognitive system as discussed so far development of communication networks and in this paper and also in [2,4,6,8]. investigate in more detail the concept of cognitive network. Cognitive networks are promising to be the major step towards efficient and automatic The cognitive process can operate in a centralized management of increasing complexity of way, spanning over a large network, or in a totally communication networks. distributed manner at a device level. In the first case, Cyclic Process in Cognitive Network. it might be too expensive to centralize all the network specific information that the cognition loop All systems that are able to adjust their functioning requires, while in the second case there might be too according to changes in their environment are based little knowledge available to pursue end-to-end on feedback information. Cognitive networks are no network goals. In reality, the deployment of the exception in this respect, so they will also use a cognitive functionality in a network will depend on control loop, also called cognition cycle [7, p. 7], the network specific problems and will be an feedback loop , context based adaptation loop . engineering decision. However, it is important that According to Thomas et al. , the loop employed the cognitive framework is designed in such way as by a cognitive network should be based on the to be modular, easily upgradeable and scalable in concept of the Observe-Orient-Decide-Act loop order to be able to accommodate existing as well as originally used in the military, augmented by next generation technologies and applications. learning and following end-to-end goals to achieve cognition. In , the loop also has a communicating The capabilities of a Cognitive Network can be capability for communicating with other loops in a highly distributed or extremely centralized. In distributed environment. general, a Cognitive network is formed of a set of 123 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 The cognition cycle as described by Mitola [7, p. 8] Loop for security features the following states: observe, orient, plan, decide, act and learn. It uses the orient module for The CycSecure application  makes use of an classifying stimuli and does not explicitly encompass incomplete cognitive loop. It uses daemons installed policies. on machines in the network that collect local information and send it to the server when polled. A Cycle management: human operator can examine and modify the network model, query and view network statistics. In , the authors investigate a cognitive agent for The system is able to generate possible attack plans wireless network selection which is designed to hide based on the information gathered from the system the complexity of the wireless environment from the and the internal knowledge base. Based on these user. The selection problem is decomposed attack plans, the human operator can decide for remedy measures to increase the security of the into four elements that enhance the agent to select system. the network which is most suitable to user preferences. First, user’s feedback that the decision Communication Requirements and research making process will be used is captured. Second, the directions available services are evaluated against learned user preferences. Third, the agent decides when to change In the history of telecommunications, development services and which new service to select based on has always been driven by humans need to user’s preferences, context and goals. communicate, i.e. reliably transmit ever increasing amount of information across increasing distances. Fourth, the value of previously unseen services is However, communication networks became predicted. Using this approach, the agent increasingly complex and more difficult to manage, continuously monitors the wireless environment and requiring increasingly specialized tools and human selects the best service according to the current operators for their maintenance, configuration and model of user preferences. However, when the user optimization.From the user’s point of view the is unsatisfied (or changes preferences), The model is necessities in the world of telecommunications, as it updated and a new selection is made to satisfy is today, are : higher bandwidth or alternative preferences.A Cognitive Resource Manager (CRM) solutions capable of accommodating the traffic . and its conceptual architecture are introduced in These necessities derive from the user’s thirst for . The CRM’s functioning is based on a cognition digital content. cycle adapted from Mitola  and aims at enabling autonomic optimization of the communication stack From the network operators’ point of view, some of as a whole, thus acting as an intelligent vertical the main necessities are: complexity, management, calibration(Fig.1). The intelligence would be based security, scalability, fault tolerance, fast integration upon methods from the field of AI. of new technologies and a good business model . The network operator has to create adequate premises for delivering the digital content. These user’s and network operators necessities are actually forming the basis for research activities currently underway in the area of cognitive networks. In general, research directions in communications can be classified in 8 broad categories: theory, signal processing, networks, software, user satisfaction, security, management and next generation protocols and architectures. In an attempt to obtain an objective big picture of the Fig 1. Open Systems Interconnection (OSI) trends in research areas as well as quantitative model estimation of the ongoing work, we used ontogeny, a semi-automatic ontology editor  to analyze the 124 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 conference proceedings of IEEE Globecom 2006 and 2007, totaling 2011 papers Artificial intelligence: Artificial intelligence is concerned with intelligent behavior in artifacts. Intelligent behavior, in turn, involves perception, reasoning, learning, communicating, and acting in complex environments .Artificial Intelligence has as one of its long-term goals the development of machines that can do these things as well as humans can, or possibly even Figure 2. Intelligent agents for Cognitive Networks better. Another goal of AI is to understand behavior whether it occurs in machines or in humans or other The starting point towards developing a CN is the animals intelligent agent (IA). This section presents existing and emerging AI techniques that can prove useful for Intelligent agents developed in a couple of streams developing agents for CNs. According to Russell and of work, among them is cybernetics [Wiener 198], Norvig [13, p. 42], an agent is central to AI. It is an cognitive psychology, Computational linguistics entity that perceives the environment through [Chomsky 1914], and adoptive control theory sensors and acts upon that environment through [Widrow & Hoff 1960], also contributed to the actuators. This is the so called ‘‘weak” definition of intellectual matrix developed by Artificial agency while ‘‘stronger” definitions take into intelligence. account functions and characteristics of the agent [14, p. 8,13, p. 42]. Among different classifications Intelligent Agents: of agents, we will consider as a reference, the one established at IBM, which uses three dimensions to Intelligent agents in Artificial intelligence react, describe agents (see Fig.6). The first dimension is plan, reason and learn in an environment more or the Agency, which determines the degree of less compatible with its abilities and goals. Here we ‘‘autonomy and authority vested in the agent”. The shall see how the actions of other agents can be second dimension is the Intelligence, which anticipated in each agents own planning, and indeed, describes the degree of reasoning and learned how an agent can even affect the actions of other behavior. Finally, the third dimension is Mobility, agents in the service of its own goals. To predict which specifies the degree to which agents travel what another agent will do , we need methods for through the network [14, p. 9]. Current networks one agent to model another ; to affect what another operate via message passing (i.e. IP packets between agent will do.There are two kinds of models used by two routers or primitives between TCP and IP) agents, iconic and feature based. An iconic model of where the receiver takes an action as a consequence the environment attempts to simulate relevant aspects of the environment; a feature-based model of the received message. This type of operation is attempts to describe the environment-perhaps by asynchronous and is characteristic to expert systems formulas in the predicate calculus. The agents that [14, p. 9,15]. This approach permitted loose coupling we deploy can use either an iconic or a feature-based of complex systems (e.g. communication networks). model of the other agents cognitive structure. And However, this approach permits the lowest degree of the other agent itself might be presumed to be using autonomy according to Fig. 1. On the Intelligence either an iconic or feature-based model. The four axis, some of the current communication systems do possibilities are shown in table 1 along with the not even reach the lowest level as they do not even modeling strategy each one provokes. allow specification of preferences (e.g. QoS specifications). In this respect, CNs are expected to enhance the level of intelligenceof current communication systemsby incorporating so called 125 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 Intelligent Agents (IAs) in the KP. On the Agency the minimal requirement for an Intelligent Agents in axis, IAs can perform actions on behalf of the user, general is to hold a model and be able to reason based on this model. These IAs (Intelligent Agents more specifically they can interact with data, )are also called knowledge-based agents. Reasoning applications or services. On the Intelligence axis, IAs can take place upon two types of knowledge: certain can hold a model(i.e. user, system, environment, (true, false and unknown) and uncertain. Reasoning etc.), perform reasoning, planning and learning. under certain knowledge is accomplished by logical These actions are exactly the same as the ones agents. In this respect, agents ‘‘can form desired from CN and can be found in the states of the representations of the world, use a process of cognition loop (see Plan, Decide, Act, Learn and [logical] inference to derive new representations Policy Fig.4). about the world, and use these new representations to deduce what to do” [13, p. 191]. Logical agents use symbolic knowledge representations, so called artificial languages, and typically first-order logic to infer new facts. These representations also support semantic querying. Agents that have incomplete or uncertain information use decision theory and are also called decision theoretic agents. These agents use knowledge representations specific for uncertain domains (i.e. full joint distributions can constitute the knowledge base) to reason. Then they perform probabilistic inference, which is the computation of posterior probabilities from the observed evidence. Networks of the future will make use of agents to Conclusions improve their performance with respect to all three axes in Fig 1. The recently emerging CN concept is promising to be the right answer to emerging challenges of the network management. In this paper we surveyed existing work on CNs. We first analyzed recent research trends in communications. We mapped existing AI techniques to the states of the cognition loop and identified challenges for research in AI from which CNs could benefit. We concluded the paper with identification of standardization activities related to or potentially benefiting from the research in the area of CNs. The discussions in this paper indicate that the way forward in developing CNs is to bring together the experts from the areas of communication networks and AI. Communication networks are faced with great complexity challenges and several AI In the case of CNs, the main improvement is techniques proved to handle complexity well. achieved with respect to the Intelligence axis. Furthermore, AI is searching for areas of Therefore, in the remainder of the section we focus applications, and communication networks are on describing utility of Intelligent Agents for these underexploited in this respect. However, due to the networks. We also emphasize the correspondence vastness in Artificial Intelligence field, we hope to between Intelligent Agents and the states of the upgrade more in terms of Cognitive Networks and cognition loop. From the intelligence point of view, other methods & tools of AI. 126 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010 References:  Nils J. Nilsson,Artificial Intelligence A New Synthesis.  R.W. Thomas, L.A. DaSilva, A.B. MacKenzie, Cognitive networks, in: Proceedings of the First IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks, Baltimore, MD, USA, November 8–11, 2006.  J. Mitola, Cognitive Radio – An Integrated Agent Architecture for Software Defined Radio, Ph.D. Dissertation, Royal Institute of Technology, Kista, Sweden, May 8, 2000  Q. Mahmoud, Cognitive Networks – Towards Self-Aware Networks, John Wiley and Sons, 2007, ISBN 9780700999.  D.D. Clark, C. Partrige, J.C. Ramming, J.T. Wroclawski, A knowledge plane for the internet, in: Proceedings of the SIGCOMM 2004, Karlsruhe, Germany, August 26–29, 2004.  R.W. Thomas, Cognitive Networks, Ph.D. Dissertation, Virginia Polytechnic and State University, Blacksburg, VA, June 16, 2007.  J. Mitola, Cognitive Radio – An Integrated Agent Architecture for Software Defined Radio, Ph.D. Dissertation, Royal Institute of Technology, Kista, Sweden, May 8, 2000.  P. Balamuralidhar, R. Prasad, A context driven architecture for cognitive nodes, Wireless Personal Communications 16 (2008) 124– 110.  FCC, ET Docket No. 04-422, Notice of Proposed Rule Making and Order, December 2004. <http://www.scribd.com/doc/112914/ Federal-Communications-Commission-FCC04422A1>.  QWL-QoS Ontology. <http://www4.ntu.edu.sg/home6/PG0487868/ OWLQoSOntology.html> (visited on August 2008).  P. Mahonen, M. Petrova, J. Riihijarvi, M. Wellens, Cognitive wireless networks: your network just became a teenager, in: Proceedings of the INFOCOM 2006, Barcelona, Spain, April 24–29, 2006.  B. Shepard, C. Matuszek, C.B. Fraser, W. Wechtenhiser, D. Crabbe, Z. Gundordu, J. Jantos, T. Hughes, L. Lefkowitz, M. Witbrock, D. Lenat, E. Larson, A knowledge-based approach to network security: applying Cyc in the domain of network risk assessment, in: Proceedings of the Innovative Applications of Artificial Intelligence Conference, Pittsburgh, PA, USA, July 9–14, 2006.  S. Russell, P. Norvig, Artificial Intelligence: A Modern Approach, second ed., Prentice Hall, 2002, ISBN 0147901262.  J. Bradshaw, Software Agents, AAAI Press/The MIT Press, 1997, ISBN 0262622412.  P. Jackson, Introduction to Expert Systems, Addison-Wesley International Computer Science Series, 1986, ISBN 0201142246. AUTHORS PROFILE I am currently employed in Jazan University, Jazan, K.S.A Department of Computer Networks. My area of Research is Artificial Intelligence, Data Mining, Network Security. I have published few papers in International Journals. I wish to do research more into these fields. 127 http://sites.google.com/site/ijcsis/ ISSN 1947-5500