Results 1  10
of
71
A tutorial on support vector machines for pattern recognition
 Data Mining and Knowledge Discovery
, 1998
"... The tutorial starts with an overview of the concepts of VC dimension and structural risk minimization. We then describe linear Support Vector Machines (SVMs) for separable and nonseparable data, working through a nontrivial example in detail. We describe a mechanical analogy, and discuss when SV ..."
Abstract

Cited by 2272 (11 self)
 Add to MetaCart
The tutorial starts with an overview of the concepts of VC dimension and structural risk minimization. We then describe linear Support Vector Machines (SVMs) for separable and nonseparable data, working through a nontrivial example in detail. We describe a mechanical analogy, and discuss when SVM solutions are unique and when they are global. We describe how support vector training can be practically implemented, and discuss in detail the kernel mapping technique which is used to construct SVM solutions which are nonlinear in the data. We show how Support Vector machines can have very large (even infinite) VC dimension by computing the VC dimension for homogeneous polynomial and Gaussian radial basis function kernels. While very high VC dimension would normally bode ill for generalization performance, and while at present there exists no theory which shows that good generalization performance is guaranteed for SVMs, there are several arguments which support the observed high accuracy of SVMs, which we review. Results of some experiments which were inspired by these arguments are also presented. We give numerous examples and proofs of most of the key theorems. There is new material, and I hope that the reader will find that even old material is cast in a fresh light.
Subalgebras of C ∗ algebras III: Multivariable operator theory
 Acta Math
, 1998
"... Abstract. A dcontraction is a dtuple (T1,..., Td) of mutually commuting operators acting on a common Hilbert space H such that ‖T1ξ1 + T2ξ2 + · · · + Tdξd ‖ 2 ≤ ‖ξ1 ‖ 2 + ‖ξ2 ‖ 2 + · · · + ‖ξd ‖ 2 for all ξ1, ξ2,..., ξd ∈ H. These are the higher dimensional counterparts of contractions. We sho ..."
Abstract

Cited by 76 (7 self)
 Add to MetaCart
Abstract. A dcontraction is a dtuple (T1,..., Td) of mutually commuting operators acting on a common Hilbert space H such that ‖T1ξ1 + T2ξ2 + · · · + Tdξd ‖ 2 ≤ ‖ξ1 ‖ 2 + ‖ξ2 ‖ 2 + · · · + ‖ξd ‖ 2 for all ξ1, ξ2,..., ξd ∈ H. These are the higher dimensional counterparts of contractions. We show that many of the operatortheoretic aspects of function theory in the unit disk generalize to the unit ball Bd in complex dspace, including von Neumann’s inequality and the model theory of contractions. These results depend on properties of the dshift, a distinguished dcontraction which acts on a new H 2 space associated with Bd, and which is the higher dimensional counterpart of the unilateral shift. H 2 and the dshift are highly unique. Indeed, by exploiting the noncommutative Choquet boundary of the dshift relative to its generated C ∗algebra we find that there is more uniqueness in dimension d ≥ 2 than there is in dimension one.
A Unifying Construction of Orthonormal Bases for System Identification
 IEEE TRANSACTIONS ON AUTOMATIC CONTROL
, 1994
"... In this paper we develop a general and very simple construction for complete orthonormal bases for system identification. This construction provides a unifying formulation of all known orthonormal bases since the common FIR and recently popular Laguerre and Kautz model structures are restrictive spe ..."
Abstract

Cited by 52 (20 self)
 Add to MetaCart
In this paper we develop a general and very simple construction for complete orthonormal bases for system identification. This construction provides a unifying formulation of all known orthonormal bases since the common FIR and recently popular Laguerre and Kautz model structures are restrictive special cases of our construction as is another construction method based on balanced realisations of all pass functions. However, in contrast to these special cases, the basis vectors in our unifying construction can have nearly arbitrary magnitude frequency response according to the prior information the user wishes to inject into the problem. We also provide results characterising the completeness properties of our bases.
Higherrank numerical ranges and compression problems
 Linear Algebra Appl
"... Abstract. We consider higherrank versions of the standard numerical range for matrices. A central motivation for this investigation comes from quantum error correction. We develop the basic structure theory for the higherrank numerical ranges, and give a complete description in the Hermitian case. ..."
Abstract

Cited by 17 (2 self)
 Add to MetaCart
Abstract. We consider higherrank versions of the standard numerical range for matrices. A central motivation for this investigation comes from quantum error correction. We develop the basic structure theory for the higherrank numerical ranges, and give a complete description in the Hermitian case. We also consider associated projection compression problems. 1.
Fitzpatrick functions and continuous linear monotone operators
, 2006
"... The notion of a maximal monotone operator is crucial in optimization as it captures both the subdifferential operator of a convex, lower semicontinuous, and proper function and any (not necessarily symmetric) continuous linear positive operator. It was recently discovered that most fundamental resul ..."
Abstract

Cited by 16 (12 self)
 Add to MetaCart
The notion of a maximal monotone operator is crucial in optimization as it captures both the subdifferential operator of a convex, lower semicontinuous, and proper function and any (not necessarily symmetric) continuous linear positive operator. It was recently discovered that most fundamental results on maximal monotone operators allow simpler proofs utilizing Fitzpatrick functions. In this paper, we study Fitzpatrick functions of continuous linear monotone operators defined on a Hilbert space. A novel characterization of skew operators is presented. A result by Brézis and Haraux is reproved using the Fitzpatrick function. We investigate the Fitzpatrick function of the sum of two operators, and we show that a known upper bound is actually exact in finitedimensional and more general settings. Cyclic monotonicity properties are also analyzed, and closed forms of the Fitzpatrick functions of all orders are provided for all rotators in the Euclidean plane.
Determinantal probability measures
, 2002
"... Abstract. Determinantal point processes have arisen in diverse settings in recent years and have been investigated intensively. We initiate a detailed study of the discrete analogue, the most prominent example of which has been the uniform spanning tree measure. Our main results concern relationship ..."
Abstract

Cited by 16 (3 self)
 Add to MetaCart
Abstract. Determinantal point processes have arisen in diverse settings in recent years and have been investigated intensively. We initiate a detailed study of the discrete analogue, the most prominent example of which has been the uniform spanning tree measure. Our main results concern relationships with matroids, stochastic domination, negative association, completeness for infinite matroids, tail triviality, and a method for extension of results from orthogonal projections to positive contractions. We also present several new avenues for further investigation, involving Hilbert spaces, combinatorics, homology,
Higherrank numerical ranges of unitary and normal matrices, preprint
, 608
"... Abstract. We verify a conjecture on the structure of higherrank numerical ranges for a wide class of unitary and normal matrices. Using analytic and geometric techniques, we show precisely how the higherrank numerical ranges for a generic unitary matrix are given by complex polygons determined by ..."
Abstract

Cited by 15 (1 self)
 Add to MetaCart
Abstract. We verify a conjecture on the structure of higherrank numerical ranges for a wide class of unitary and normal matrices. Using analytic and geometric techniques, we show precisely how the higherrank numerical ranges for a generic unitary matrix are given by complex polygons determined by the spectral structure of the matrix. We discuss applications of the results to quantum error correction, specifically to the problem of identification and construction of codes for binary unitary noise models. 1.
A CartanHadamard Theorem for BanachFinsler Manifolds
 Proceedings of the Conference on Geometric and Combinatorial Group Theory, Part II (Haifa, 2000), Geom. Dedicata 95
, 2002
"... . In this paper we study Banach{Finsler manifolds endowed with a spray which have seminegative curvature in the sense that the corresponding exponential function has a surjective expansive dierential in every point. In this context we generalize the classical theorem of Cartan{ Hadamard, saying t ..."
Abstract

Cited by 14 (1 self)
 Add to MetaCart
. In this paper we study Banach{Finsler manifolds endowed with a spray which have seminegative curvature in the sense that the corresponding exponential function has a surjective expansive dierential in every point. In this context we generalize the classical theorem of Cartan{ Hadamard, saying that the exponential function is a covering map. We apply this to symmetric spaces and thus obtain criteria for Banach{Lie groups with an involution to have a polar decomposition. Typical examples of symmetric Finsler manifolds with seminegative curvature are bounded symmetric domains and symmetric cones endowed with their natural Finsler structure which in general is not Riemannian. Introduction Let M = G=K be a nitedimensional noncompact Riemannian symmetric space, where K is the group of xed points of an involution on G . Then G has a polar decomposition in the sense that the decomposition g = k + p of its Lie algebra into the eigenspaces of the involution d(1) leads to a di...
Assche, Perturbation of orthogonal polynomials on an arc of the unit circle
 J. Approx. Theory
, 1995
"... Abstract. Orthogonal polynomials on the unit circle are completely determined by their reflection coefficients through the Szegő recurrences. We assume that the reflection coefficients converge to some complex number a with 0 < a  < 1. The polynomials then live essentially on the arc { eiθ: α ≤ θ ..."
Abstract

Cited by 11 (2 self)
 Add to MetaCart
Abstract. Orthogonal polynomials on the unit circle are completely determined by their reflection coefficients through the Szegő recurrences. We assume that the reflection coefficients converge to some complex number a with 0 < a  < 1. The polynomials then live essentially on the arc { eiθ: α ≤ θ ≤ 2π −α} where cos α def √ = 1 − a  2 with α ∈ (0, π). We analyze the 2 orthogonal polynomials by comparing them with the orthogonal polynomials with constant reflection coefficients, which were studied earlier by Ya. L. Geronimus and N. I. Akhiezer. In particular, we show that under certain assumptions on the rate of convergence of the reflection coefficients the orthogonality measure will be absolutely continuous on the arc. In addition, we also prove the unit circle analogue of M. G. Krein’s characterization of compactly supported nonnegative Borel measures on the real line whose support contains one single limit point in terms of the corresponding system of orthogonal polynomials. 1.
Factorization of Almost Periodic Matrix Functions of Several Variables and Toeplitz Operators
 Math.Notes45 (1989), no. 5–6, 482–488. MR 90k:47033
"... this paper. We let (AP ..."