A. Kuh

Imperial College London, London, ENG, United Kingdom

Are you A. Kuh?

Claim your profile

Publications (51)44.98 Total impact

  • M. Uddin, A. Kuh, A. Kavcic
    [Show abstract] [Hide abstract]
    ABSTRACT: The objective of this paper is to find numerical upper bounds on the optimal solution to the sensor placement problem. Given noisy measurements and knowledge of the state correlation matrix, the sensor placement problem can be formulated as an integer programming problem using a linear minimum mean squared error estimator. Since finding the optimal placements of a fixed number of sensors in a large network is computationally infeasible, finding bounds for the optimal solution is a fundamental task. In this paper we present a family of nested bounds using matrix pencils and their generalized eigenvalues that upper bound the optimal performance. In the analysis we consider nodes that we want to place sensors and other nodes where we cannot or do not want to place sensors. Finally we compare the upper bounds with the optimal solution using simulations on a 5 by 5 grid network.
    Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on; 01/2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: Adaptive online algorithms for simultaneously extracting nonlinear eigenvectors of kernel principal component analysis (KPCA) are developed. KPCA needs all the observed samples to represent basis functions, and the same scale of eigenvalue problem as the number of samples should be solved. This paper reformulates KPCA and deduces an expression in the Euclidean space, where an algorithm for tracking generalized eigenvectors is applicable. The developed algorithm here is least mean squares (LMS)-type and recursive least squares (RLS)-type. Numerical example is then illustrated to support the analysis.
    Acoustics, Speech and Signal Processing (ICASSP), 2012 IEEE International Conference on; 01/2012
  • [Show abstract] [Hide abstract]
    ABSTRACT: A novel class of complex valued kernel least mean square (CKLMS) algorithms is introduced with the aim to provide physical meaning to the mapping between the primal and dual space termed the independent CKLMS (iCKLMS). The general class of CKLMS algorithms is also extended in the widely linear sense to develop online kernel algorithms suitable for the processing of general complex valued signals, both circular and noncircular. The so-introduced augmented complex kernel least mean square (ACKLMS) algorithms are verified on adaptive prediction of nonlinear and nonstationary complex wind signals.
    Sensor Array and Multichannel Signal Processing Workshop (SAM), 2012 IEEE 7th; 01/2012
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper considers the placement of m sensors at n > m possible locations. Given noisy observations, knowledge of the state correlation matrix, and a mean square error criterion, the problem can be formulated as an integer programming problem. The solution for large m and n is infeasible, requiring us to look at approximate algorithms. Using properties of matrices, we come up with lower and upper bounds for the optimal solution performance. We also formulate a greedy algorithm and a dynamic programming algorithm that runs in polynomial time of m and n. Finally, we show through simulations that the greedy and dynamic programming algorithms very closely approximate the optimal solution. The sensor placement problem has many energy applications where we are often confronted with limited resources. Some examples include where to place environmental sensors for an area where there are large amounts of distributed solar PV and where to place grid monitors on an electrical distribution microgrid.
    Smart Grid Communications (SmartGridComm), 2012 IEEE Third International Conference on; 01/2012
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a new probabilistic approach of the real-time state estimation on the micro-grid. The grid is modeled as a factor graph which can characterize the linear correlations among the state variables. The factor functions are defined for both the circuit elements and the renewable energy generation. With the stochastic model, the linear state estimator conducts the Belief Propagation algorithm on the factor graph utilizing real-time measurements from the smart metering devices. The result of the statistical inference presents the optimal estimates of the system state. The new algorithm can work with sparse measurements by delivering the optimal statistical estimates rather than the solutions. In addition, the proposed graphical model can integrate new models for solar/wind correlation that will help with the integration study of renewable energy. Our state-of-art approach provides a robust foundation for the smart grid design and renewable integration applications.
    Neural Networks (IJCNN), The 2011 International Joint Conference on; 09/2011
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The most popular method used in traditional power system state estimation is the Maximum Likelihood Estimation (MLE). It assumes the state of the system is a set of deterministic variables and determines the most likely state via error included interval measurements. In the distribution system, the measurements are often too sparse to fulfill the system observability. Instead of introducing pseudo-measurements, we propose a Belief Propagation (BP) based distribution system state estimator. This new approach assumes that the system state is a set of stochastic variables. With a set of prior distributions, it calculates the posterior distributions of the state variables via real-time sparse measurements from both traditional measurements and the high resolution smart metering data. In this paper we discuss the step-by-step method of applying the BP algorithm on the distribution system state estimation problem. Our approach provides a seamless connection from the monitoring of transmission system to the feeder circuit, thus filling in the gap between the traditional energy management system (EMS) and the micro-grid customer level optimization. Furthermore, the proposed state estimator can not only be applied to the multi-level electrical coupled grid, but also accommodate the spatial-temporal model for the correlated distributed renewable energy resources. It provides a way of integrating the distributed renew able energy management system into the Smart-Grid Distribution Management System (DMS) and automated substations.
    IEEE Computational Intelligence Magazine 09/2011; · 4.63 Impact Factor
  • N. Kowahl, A. Kuh
    [Show abstract] [Hide abstract]
    ABSTRACT: The adoption of smart grid technologies will allow for more distributed generation of energy and for residential and commercial users of electricity to make intelligent decisions about energy usage. In previous research by Livengood and Larsen, a stochastic dynamic programming problem is formulated for a micro-scale smart grid system. A mathematical model of energy usage is developed where the goal is to optimize a finite horizon cost function reflecting both the cost of electricity and comfort/lifestyle. This paper extends this work by assuming key models and forecasts are unknown and implicitly learned via the softmax algorithm with neighborhood updating. The algorithm implements approximate dynamic programming with a goal of reducing dependancies on models and forecasting while achieving good performance. Simulations are conducted using the softmax algorithm showing that the solution approaches the optimal dynamic programming algorithm solution.
    Neural Networks (IJCNN), The 2010 International Joint Conference on; 08/2010
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The second order Taylor series expansion (TSE) of scalar functions of complex matrices is explored in order to provide a new tool for gradient-based optimisation in the complex domain. The expansion is provided both in the augmented real and complex matrix spaces, as well as the multidimensional complex domain. The duality (isomorphism) between the augmented spaces is established and consequently the relation of the first- and second-order terms (gradient and Hessian) of the TSE in these spaces are introduced. Finally, a study of the trade-off between performance and computational complexity of algorithms for the estimation of complex sources in the two augmented spaces is performed.
    Green Circuits and Systems (ICGCS), 2010 International Conference on; 07/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we discuss a new set of nonlinear adaptive filters based on kernel methods and compare them to the least mean square (LMS) and recursive least squares (RLS) adaptive filters. In recent years a new class of nonlinear kernel adaptive filters have been developed that tradeoff performance for complexity including the Kernel LMS (KLMS) and Kernel RLS (KRLS) algorithms. Earlier work discussed a complex augmented implementation of the kernel algorithms. This paper continues this discussion and compares the performance and complexity of the algorithms for wind time series prediction.
    01/2010;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Smart grid envisions the potential to manage diverse energy resources and enable a future self-dispatch and self-healing grid. This would first require the micro-grid visibility of node behavior (i.e. electrical parameters). In this paper we propose a novel approach to construct a stochastic model that makes global inference on every node at the micro-grid level. The micro-grid system can be modeled as a factor graph addressing proper correlation functions including distributed re-newable generation correlation. We conduct statistical inference on the factor graph using Belief Propagation (BP) algorithm. The purpose is that given incomplete measurements, marginal probability distribution for unmetered node behavior can be derived. Simulation of the BP algorithm is performed on a simplified micro-grid model with linear local correlations. The results demonstrate that loopy BP can converge to optimal state estimates efficiently.
    01/2010;
  • Source
    A. Kuh, D. Mandic
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper combines complex signal processing with kernel methods for applications in wind prediction. Specifically, we consider developing least squares kernel algorithms for both complex data and augmented complex data. The augmented complex kernel algorithms have advantages over complex kernel algorithms in both the areas of performance and complexity. Use of kernels also allow implementation of nonlinear algorithms by working in the dual space. We apply our algorithm to wind series time prediction and show that our augmented complex algorithms outperform other complex least square algorithms.
    Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on; 05/2009
  • Source
    Anthony Kuh, Danilo Mandic
    [Show abstract] [Hide abstract]
    ABSTRACT: APSIPA ASC 2009: Asia-Pacific Signal and Information Processing Association, 2009 Annual Summit and Conference. 4-7 October 2009. Sapporo, Japan. Oral session: Signal Processing Theory and Methods II (7 October 2009). This paper discusses some recent advances in the development of nonlinear adaptive filtering. Specifically, it studies online kernel adaptive filters. We study the performance and complexity of a suite of kernel online algorithms from kernel recursive least square subspace (KRLSS) algorithms to kernel least mean square (KLMS) algorithms. A key to the kernel algorithms is that updating is done in the dual space via evaluation of kernels, the number of support vectors is controlled by using an information criterion, and the number of information vectors is controlled for KRLSS These algorithms have advantages in that they are nonlinear filters that are relatively easy to implement and have a number of parameters that can be adjusted to tradeoff performance for complexity.
    01/2009;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: a b s t r a c t This paper presents a novel approach for the simultaneous modelling and forecasting of wind whereby the wind field is considered as a vector of its speed and direction components in the field of complex numbers C. To account for the intermittency and coupling of wind speed and direction, we propose to use the recently introduced framework of augmented complex statistics. The augmented complex least mean square (ACLMS) algorithm is introduced and its usefulness in wind forecasting is analysed. Simulations over different wind regimes support the approach.
    Renewable Energy. 08/2008; 34(1).
  • Source
    Chaopin Zhu, A. Kuh
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we apply distributed kernel regression methods to perform sensor network localization. This follows up on earlier work where a centralized kernel regression algorithm was considered to perform localization. Here we examine the tradeoffs between using distributed algorithms versus centralized algorithms in terms of communication costs, computational costs, and performance of the estimate. Simulation results demonstrate that distributed methods work well with comparable performance to centralized algorithms with less communication costs
    Acoustics, Speech and Signal Processing, 2007. ICASSP 2007. IEEE International Conference on; 05/2007
  • Source
    Anthony Kuh, Philippe De Wilde
    [Show abstract] [Hide abstract]
    ABSTRACT: In this letter, we comment on "Pruning Error Minimization in Least Squares Support Vector Machines" by B. J. de Kruif and T. J. A. de Vries. The original paper proposes a way of pruning training examples for least squares support vector machines (LS SVM) using no regularization (-gamma = infinity). This causes a problem as the derivation involves inverting a matrix that is often singular. We discuss a modification of this algorithm that prunes with regularization (gamma finite and nonzero) and is also computationally more efficient.
    IEEE Transactions on Neural Networks 04/2007; 18(2):606-9. · 2.95 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we study an evolving email network model first introduced by Wang and De Wilde, to the best of our knowledge. The model is analyzed by formulating the network topology as a random process and studying the dynamics of the process. Our analytical results show a number of steady state properties about the email traffic between different nodes and the aggregate networking behavior (i.e., degree distribution, clustering coefficient, average path length, and phase transition), and also confirm the empirical results obtained by Wang and De Wilde. We also conducted simulations confirming the analytical results. Extensive simulations were run to evaluate email traffic behavior at the link and network levels, phase transition phenomena, and also studying the behavior of email traffic in a hierarchical network. The methods established here are also applicable to many other practical networks including sensor networks and social networks.
    Physical Review E 11/2006; 74(4 Pt 2):046109. · 2.31 Impact Factor
  • Source
    Anthony Kuh, Chaopin Zhu, Danilo Mandic
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper considers the sensor network localization problem using signal strength. Unlike range-based methods signal strength information is stored in a kernel matrix. Least squares regression methods are then used to get an estimate of the location of unknown sensors. Locations are represented as complex numbers with the estimate function consisting of a linear weighted sum of kernel entries. The regression estimates have similar performance to previous localization methods using kernel classification methods, but at reduced complexity. Simulations are conducted to test the performance of the least squares kernel regression algorithm. Finally, the paper discusses on-line implementations of the algorithm, methods to improve the performance of the regression algorithm, and using kernels to extract other information from distributed sensor networks.
    10/2006: pages 1280-1287;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: A novel method for the assessment of the qualitative performance of machine learning algorithms is proposed. This is achieved by a modification of the recently proposed "delay vector variance" (DVV) method for the signal modality characterisation. Based on the local predictability in phase space we propose to employ the scatter diagram of DVV features in order to gauge the changes in signal nature after being processed by machine learning algorithms. A set of comprehensive simulations on representative data sets supports the analysis.
    Machine Learning for Signal Processing, 2006. Proceedings of the 2006 16th IEEE Signal Processing Society Workshop on; 10/2006
  • Chaopin Zhu, A. Kuh
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we apply complex least squares kernel subspace methods to the problem of ad hoc network localization. We use Gaussian kernels and a neighborhood kernel to estimate the locations of mobile nodes. Our algorithms do not require preprocessing of raw data like other statistical methods. Furthermore they use one-step regression directly, instead of existing two-stage classification methods, and work on a fairly small subset of training data. These salient features allow our algorithms to successfully solve the dynamic localization problem with low communication and computational costs. Simulation of ad hoc networks with random node movement demonstrates the success of the algorithms. The methods and algorithms can also be applied in other applications like target tracking and sensor data representation
    Information Theory, 2006 IEEE International Symposium on; 08/2006
  • A. Kuh, D. Mandic
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper considers sequential detection problems where we learn from sets of training sequences. The sufficient statistics can be learned quickly using a least squares temporal difference (TD) learning algorithm. This algorithm converges much quicker than previously applied TD learning algorithms. The algorithm can easily be implemented in an on-line manner and can also be applied to more complicated decentralized detection problems
    Acoustics, Speech and Signal Processing, 2006. ICASSP 2006 Proceedings. 2006 IEEE International Conference on; 06/2006

Publication Stats

218 Citations
44.98 Total Impact Points

Institutions

  • 2005–2010
    • Imperial College London
      • Department of Electrical and Electronic Engineering
      London, ENG, United Kingdom
  • 1989–2010
    • University of Hawaiʻi at Mānoa
      • Department of Electrical Engineering
      Honolulu, HI, United States
  • 2007
    • Honolulu University
      Honolulu, Hawaii, United States
  • 1989–1998
    • Hawaii Pacific University
      Honolulu, Hawaii, United States
  • 1996
    • Dalian University of Technology
      Lü-ta-shih, Liaoning, China
  • 1992
    • Stanford University
      Palo Alto, California, United States