Session 2

From ICE PhD Wiki
Jump to: navigation, search

Mobile networking / Distributed sensing

Ad-hoc mobile networking for adaptive environments

Mobile ad hoc networks (MANETs) are distributed systems consisting of wireless mobile nodes that can freely and dynamically self-organize into arbitrary ad-hoc network topologies. This technology allows people and devices to internetwork in areas with no pre-existing communication infrastructure. A good example would be disaster recovery environment. MANET is an autonomous set of mobile nodes that communicate over bandwidth constrained wireless links and cannot rely on centralized and organized connectivity. Since the nodes are mobile, the network topology may change rapidly and unpredictably over time. All network activity must be executed by the nodes themselves, i.e., routing functionality will be incorporated into mobile nodes. Ad hoc networks maximize total network throughput by using all available nodes for routing and forwarding. Establishing survivable, efficient, dynamic communication for emergency/rescue operations, disaster relief efforts, and military networks is a main motivation for research in MANETs.

  • Applications areas:
    • Personal area networking (cell phone, laptop, ear phone, wrist watch)
    • Military environments (soldiers, tanks, planes)
    • Civilian environments (taxi cab network, meeting rooms, sports stadiums, boats, UAVs)
    • Environmental, csarch, rescue operations and fire fighting

  • Key challenges:
    • Ad-hoc routing,
    • Ad-hoc (re-)configuration,
    • Ad-hoc management?

  • More open questions:
    • Different approaches to choose(i.e. REactive vs PROactive routing)
      • Challenges imposed by networks' dynamics
      • Power consumption, estimation, data/power efficiency

  • Related ICE Projects:
    • Project on related topic [related student]

  • References:
    • Imrich Chlamtac, Marco Conti, Jennifer J.-N. Liu: Mobile ad hoc networking: imperatives and challenges. Ad Hoc Networks 1(1): 13-64 (2003) [paper]

Internet of things for internet of people

The “Internet of Things” (IoT) provides connectivity for anyone at any time and place to anything at any time and place. With the advancement in technology, we are moving towards a society, where everything and everyone will be connected. The IoT is considered as the future evaluation of the Internet that realizes machine-to-machine (M2M) learning. The basic idea of IoT is to allow autonomous and secure connection and exchange of data between real world devices and applications. The IoT links real life and physical activities with the virtual world.

The IoT can be considered as the “social network of objects” where objects autonomously interact and communicate useful information, take decisions and invoke actions. The objects have certain unique features and are uniquely identifiable and accessible to the Internet. These physical objects are equipped with radio-frequency identification (RFID) tags or other identification code bars that can be sensed by the smart devices.

  • Possible Future Applications areas:
    • Medical applications
    • Agriculture application
    • Intelligent transport system design
    • Design of smart cities
    • Design of smart homes
    • Industry applications
    • Prediction of natural disasters
    • Water Scarcity monitoring
    • Smart metering and monitoring
    • Smart Security
  • Key challenges:
    • Security
    • Privacy and Data protection
    • Control of Critical Global Resources
    • Identity Management, Naming and Interoperability
    • Fostering Innovation
    • Spectrum
    • Standardization
  • Active European Research Projects
    • European Research Cluster on the Internet of Things (IERC), FP7 project 2009.
    • The Internet of Things Architecture (IoT-A), FP7 European project, Sept. 2010 – Sept .2013.
    • The IoT@Work, FP7 European project June 2010 - June 2013.
    • The Internet of Things Initiative, (IoT-i), FP7 European project , Sept. 2010 – August 2012.
    • European Internet of Things Forum, (, 2nd IoT week organized in Venice, 19-21 June, 2012.
  • References:
    • L. Coetzee and J. Eksteen, "The Internet of Things – Promise for the Future? An Introduction", Conference Proceedings IST-Africa, South Africa, May 2011 [Paper]
    • T. Lu and W. Neng, "Future Internet: The Internet of Things", 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE), China, August 2010 [Paper]
    • W. Miao, L. Ting-Jie, L. Fei-Yang, et-al, "Research on the architecture of Internet of things", 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE), Beijing, China, August 2010 [Paper]
    • H. Zhihua, "The research of several key question of Internet of Things", International Conference on Intelligence Science and Information Engineering (ISIE), Huanggang, China, August 2011 [Paper]

Resource-constrained interactive pattern-learning systems

  • Resource Coordination in Distributed Sensing
    • Distributed sensing platform consists of several sensor nodes capable of sensing, processing and communicating with other node. It generally forms the sensor networks. Sensor nodes are battery operated and very much resource (energy, computing performance, memory, avialable information, communication capabilities and processing functionality) sensitive. Effective resource coordination of available resources is an important factor in sensor networks.
    • Resource coordination helps to dyanmically adjust the resources in a way that the performed application by the nodes is resource efficient. There are several reasons for dynamically maintaining the resources: for example to increase the lifetime of a network, to overcome the problem of resource deficit especially when sensors exhibit failures, to maintain the runtime adaptation of resources and to determine the best allocation of tasks to resources.
  • State of the Art and Different Approaches:
    • Constraint Satisfaction Based: Constraint satisfaction is a formalism that has been used to model a large class of problems with applications in engineering design, planning, scheduling, resource allocation and fault diagnosis.
    • Utility Based: Utility-based technique is a valuable tool for resource management in wireless sensor networks (WSN). The benefits of this technique include the ability to take application utilities into account and to dynamically adapt to the highly variable environment of tactical WSN.
    • Predicate Logic Based: Rule or predicate help to configure the wireless sensor networks, where certain functions must be automatically assigned to sensor nodes, such that the properties of a sensor node (e.g. remaining energy, network neighbors) match the requirements of the assigned function. Based on the assigned roles, sensor nodes may adapt their behavior accordingly, establish cooperation with other nodes, or may even download specific code for the selected role.
    • Reinforcement Learning Based: Reinforcement learning helps to enable autonomous self learning/adaptive applications with inherent support for efficient resource/task management. It is the process by which an agent improves in an environment via experience. It does not need a model of its environment and can be used on line. It is simple, demands minimal computational resources.
    • Market or Auction Based: An auction is a decentralized market mechanism for allocating resources. The essence of an auction is a game, where the players are the bidders, the strategies are the bids and both allocation and payments are function of the bids. One well known auction is the Vickery-Clarke-Groves (VCG) auction, which requires gathering global information from the network and performing centralized computations.
  • Applications and challenges:
    • Increasing the lifetime of the network performing for performing applications such as obejct tracking, routing, area monitoring etc.
    • Overcoming the problem of resource deficit especially when sensors exhibit failures.
    • Maintaining the runtime adaptation of resources.
    • Determining the best allocation of tasks to resources.
    • Balancing the work load in the network.
    • Discovering the right task to perform at each time step.
  • ICE Perspective:
    • Cognitive system processes knowledge and makes decisions based on it. A cognitive system is one that can perceive the environment and adapts to it, can make intelligent decisions based on its knowledge that effect changes in the environment, can self-manage, and self-heal.
  • Ongoing ICE Projects:
    • Resource Coordination in Networked Embedded Systems Muhidul Khan
    • Energy aware networking for home environment Rafiullah Khan
    • Collaboration in Networked Unmanned Aerial Vehicles Asif Khan
  • References:
    • K.Shah and M. Kumar, Distributed Independent Reinforcement Learning (DIRL) Approach to Resource Management in Wireless Sensor Networks, Proc. IEEE Mobile Adhoc and Sensor Systems (Mass 07), 1-9, 2007.
    • R. Sutton and A. Barto, Reinforcement Learning: An Itroduction, MIT Press, ISBN 9-262-19398-1.
    • M.I.Khan, B.Rinner, Resource Coordination in Wireless Sensor Networks by Cooperative Reinforcement Learning, Proc. 10th IEEE International Conference on Pervasive Computing and Communications (PerCom 2012), 901-906, 2012.
    • N. Edalat, W. Xiao, N. Roy, S. Das and M. Motani, Combinatorial Auction Based Task Allocation in Multi Application Wireless Sensor Networks, IFIP International Conference on Embedded and Ubiquitous Computing, 174-181, 2011.
    • J. Huang, Z. Han, M. Chiang and H. Poor, Auction based resource allocation for cooperative communications, IEEE Journal on selected areas in communication, vol. 26, no. 7, September 2008.
    • J.M.Vidal, Multiagent Coordination Using a Disributed Combinatorial Auction, AAAI Workshop on Auction Mechanisms for Robot Coordiantion, 1-6, 2006.
    • C. Frank and K. Romer, Algorithms for role assignments in wireless sensor networks, In proceedings of the 3rd international conference on Embedded networked sensor systems, 2005.
    • R. Dechter and D. Frost, Backtracking algorithms for constraint satisfaction problems, Technical Report, Information and Computer Science Department, UC Irvine, 1999.
    • B. Krishnamachari, S. Wicker, R. Bejar and C. Fernandez, On the complexity of distributed self configuration in wireless networks, Telecommunication Systems 22:1-4, 33-59, 2003.

Machine Learning and Pattern Recognition

  • Definition:

Machine Learning (ML) is a branch of Artificial Intelligence (AI) that studies and develops the methodologies to teach computers how to operate or behave based on empirical data. In other words, it aims to provide machines certain ‘Intelligence’ and generalization capability in response to inputs and external information. It has been widely applied to several fields of study such as Computer Vision, telecommunications, and medical diagnosis.

  • Process Pipeline:

A conventional ML System can be described in two main blocks: First a Training Phase which consists of feeding the system with samples in order to find an optimal model that learns the complex relationships within the data. And second the Prediction Phase that classifies novel patterns using the learned model. There exists several approaches to ML Systems ranging from deterministic threshold-based models to more advanced probabilistic methods such as Bayesian Networks.

  • Applications:

The application of ML Systems in several ICE areas such as in Distributed Sensing and Mobile Networking is becoming increasingly common. Mainly as a consequence of the vast amounts of mobile networks in development and the need for fast and automated systems to control and monitor them. For instance: Body Sensor Networks, Cognitive Radio, Ad-Hoc Networks, Network Intrusion detection. Some of these applications are detailed below:

    • Ad-Hoc Networks and Energy-Aware Routing: In Ad-Hoc network configurations traditional routing protocols cannot be applied successfully as they might reduce the QoS and increase the overall energy consumption. Current research at the network layer routing protocol is being made to solve this issue by implementing ML algorithms. The network should be able to learn the characteristics of the environment and adapt itself accordingly. For instance: Reinforcement Learning methods such as Q-Learning have been applied to Wireless Sensor Networks and are well suited for solving distributed network problems such as routing.
    • Network Intrusion: Nowadays network security and the need of continuous supervision has become a relevant subject of research. Human operators cannot cope with the amount of work needed for this activity and they are being replaced by autonomous systems. As an example, network vulnerabilities can be circumvented by using ML methods such as K-Nearest neighbour classifiers for detecting intruder system calls. These methods also intends to adapt as network changes and evolve with new user behaviour and data.
    • Learning Engine for Cognitive Radio: Most Cognitive Radio (CR) technologies use policy-based radios based on a fixed set of rules that define their behaviour in different conditions and scenarios. New CR approaches consider the adaptation of ML techniques that modify their reasoning engine using past experience information to improve the radio performance on making decisions about future actions. For example, Neural Networks and Genetic Algorithms have been proposed for improving the maximization of the CR channel capacity.
    • Learning on constrained distributed sensing: Some distributed sensing configurations are limited regarding bandwidth, for instance the monitoring of certain manufacturing processes and climate changes in vast areas. For reducing the amount of data traffic it is possible to perform onsite learning on each sensor location and avoid transferring redundant data o other nodes. The application of Support Vector Machines has been used on this by learning in a distributed manner. The output of these algorithms is transferred instead of the raw sensor data reducing to a great extent the required bandwidth.
    • Analysis data from Body Sensor Networks: Body sensor networks are being employed for the analysis of medical conditions such as in patients with Parkinson’s Disease. Data is generally collected from a group of sensors attached to the body (e.g. accelerometers, gyroscopes, heart rate monitors). They provide a considerable amount of information in terms of time domain signals that needs to be analysed. Different patients generate different patterns and data can vary significantly within them. Some studies are required to find common indications regarding the patients’ health conditions. ML methods such as Support Vector Machines has been used for classification of patient activities, detection of ON-OFF state and Freezing of Gait.

  • References:
    • P. Nurmi. "Reinforcement Learning for Routing in Ad Hoc Networks". 5th International Symposium on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks and Workshops, 2007. WiOpt 2007. pp.1-8, 16-20 April 2007. [[1]]
    • M. Govindarajan and R.M. Chandrasekaran. "Intrusion detection using k-Nearest Neighbor". First International Conference on Advanced Computing, 2009. ICAC 2009. pp.13-20, 13-15 Dec. 2009. [[2]]
    • C. Clancy, J. Hecker, E. Stuntebeck and T. O'Shea. "Applications of Machine Learning to Cognitive Radio Networks". IEEE Wireless Communications vol.14, no.4, pp.47-52, August 2007. [[3]]
    • P.A. Forero, A. Cano, and G.B. Giannakis. "Consensus-Based Distributed Support Vector Machines". J. Mach. Learn. Res. 11 (August 2010), 1663-1707. [[4]]
    • S. Patel, K. Lorincz, R. Hughes, N. Huggins, J. Growdon, D. Standaert, M. Akay, J. Dy, M. Welsh and P. Bonato. "Monitoring Motor Fluctuations in Patients With Parkinson's Disease Using Wearable Sensors". IEEE Transactions on Information Technology in Biomedicine, vol.13, no.6, pp.864-873, Nov. 2009. [[5]]

Distributed algorithms for sensor networks

The recent advances in wired and wireless technology lead to the emergence of large-scale networks such as Internet, Mobile ad-hoc networks and Wireless sensor networks. The advances gave rise to new network applications including distributed network operations including resource allocation, coordination, learning, and estimation.

Wireless sensor networks are capable of collecting an enormous amount of data over space and time. Often, the ultimate objective is to derive an estimate of a parameter or function from these data. A general class of distributed algorithms eliminate the need to transmit raw data to a central point. This can provide significant reductions in the amount of communication and energy required to obtain an accurate estimate.

  • Main challenges of distributed environments:
    • Impossibility of centralized network architecture
    • Size of the network / Proprietary issues
    • Network dynamics
    • Mobility of the network
    • The agent spatio-temporal dynamics
    • Network connectivity structure is varying in time

  • Research Areas in Distributed algorithms are:
    • Consensus algorithms
    • Information diffusion
    • Opinion dynamics and Stability

  • Related ICE Projects:

  • References:
    • Robustness of Self-Organizing Consensus Algorithms: Initial Results from a Simulation-Based Study, Gogolev A. and Bettstetter C., in proc. IWSOS-2012. [6]
    • Robustness and Dependability of Self-Organising Systems – A Safety Engineering Perspective, Di Marzo Serugendo, G. The 11th International Symposium on Stabilization, Safety and Security of Distributed Systems (SSS 2009), Lyon, France, November 2009.[7]
    • Efficcient system-wide coordination in noisy environments, Moreira, A. A., Mathur, A., Diermeier, D., Amaral, L. In: Proc. National Academy of Sciences of the USA, vol. 101, no. 33, pp. 12085{12090, (2004)[8]

Students: Md. Muhidul Islam Khan, Rafiullah Khan, Alexander Gogolev, Jorge Luis Reyes Ortiz, Asif Khan, Venkata Satya Rajendra Pathuri Bhuvana

Supervisor: Prof. Bernhard Rinner