Results 1 
4 of
4
Generalized binary search
 In Proceedings of the 46th Allerton Conference on Communications, Control, and Computing
, 2008
"... This paper addresses the problem of noisy Generalized Binary Search (GBS). GBS is a wellknown greedy algorithm for determining a binaryvalued hypothesis through a sequence of strategically selected queries. At each step, a query is selected that most evenly splits the hypotheses under consideratio ..."
Abstract

Cited by 30 (0 self)
 Add to MetaCart
This paper addresses the problem of noisy Generalized Binary Search (GBS). GBS is a wellknown greedy algorithm for determining a binaryvalued hypothesis through a sequence of strategically selected queries. At each step, a query is selected that most evenly splits the hypotheses under consideration into two disjoint subsets, a natural generalization of the idea underlying classic binary search. GBS is used in many applications, including fault testing, machine diagnostics, disease diagnosis, job scheduling, image processing, computer vision, and active learning. In most of these cases, the responses to queries can be noisy. Past work has provided a partial characterization of GBS, but existing noisetolerant versions of GBS are suboptimal in terms of query complexity. This paper presents an optimal algorithm for noisy GBS and demonstrates its application to learning multidimensional threshold functions. 1
Measures for Tracing Convergence of Iterative Decoding Algorithms
 in Proc. 4th IEEE/ITG Conf. on Source and Channel Coding
, 2002
"... We study the convergence behavior of turbo decoding, turbo equalization, and turbo bitinterleaved coded modulation in a unified framework, which is to regard all three principles as instances of iterative decoding of two serially concatenated codes. There is a collection of measures in the recent l ..."
Abstract

Cited by 28 (5 self)
 Add to MetaCart
We study the convergence behavior of turbo decoding, turbo equalization, and turbo bitinterleaved coded modulation in a unified framework, which is to regard all three principles as instances of iterative decoding of two serially concatenated codes. There is a collection of measures in the recent literature, which trace the convergence of iterative decoding algorithms based on a single parameter. This parameter is assumed to completely describe the behavior of the softin softout decoders being part of the iterative algorithm. The measures observe different parameters and were originally applied to different types of decoders. In this paper, we show how six of those measures are related to each other and we compare their convergence prediction capability for the decoding principles mentioned above. We observed that two measures predict the convergence very well for all regarded decoding principles and others suffer from systematic prediction errors independent of the decoding principle.
nport resistive network synthesis from prescribed sensitivity coefficients
 IEEE Truns. Circuits Syst
, 1975
"... AbstractAn investigation of the relationship between the sensitivity coefficients and the Kmatrix of nport networks is presented. The results available on the Kmatrix and the adjoint network approach for sensitivity computations will form the basis of the discussions. THE PMATRIX ONSIDER a resi ..."
Abstract

Cited by 1 (0 self)
 Add to MetaCart
AbstractAn investigation of the relationship between the sensitivity coefficients and the Kmatrix of nport networks is presented. The results available on the Kmatrix and the adjoint network approach for sensitivity computations will form the basis of the discussions. THE PMATRIX ONSIDER a resistive nport network N. Let 4 = C {gi} denote the column matrix of edge conductances of N. We shall denote by Jo. = {gio} the nominal value of @., The sensitivity coefficient matrix P = [pii] with respect to the Ymatrix of N will be defined as follows: Pij = 218 =Oo I where yii is the shortcircuit drivingpoint admittance across port i of N. Pij will be referred to as the sensitivity coefficient of yii with respect to the conductance gj. It may be seen that the matrix P is of order it x e, where e is the number of edges in N. Without any loss of generality the graph of the network N may be assumed to be complete by permitting edges with zero admittances. It follows from the results of [l], [2] that where vkj (a,‘) is th e voltage across conductance gk when port j(i) is excited with a source of unit voltage and all the other ports are shortcircuited. This is illustrated in Fig. 1. Hence we get Pik IO = 80 = Cvk’12 (3) as a consequence of the no amplification property of resistive networks pik I 1. Let the port configuration T of N be in p parts Tl, T,, * *., Tp. Let the set of vertices in Ti be denoted by l$. Let To be a tree of N such that T c To. We shall denote by qi thefcutset of N with respect to the branch of T corresponding to the port i. If port i is in Tk, it may be seen from (3)