John Hannah

Bio: John Hannah is an academic researcher from University College Dublin. The author has an hindex of 1, co-authored 1 publications receiving 48 citations.

On Copositive Programming and Standard Quadratic Optimization Problems

Abstract: A standard quadratic problem consists of finding global maximizers of a quadratic form over the standard simplex. In this paper, the usual semidefinite programming relaxation is strengthened by replacing the cone of positive semidefinite matrices by the cone of completely positive matrices (the positive semidefinite matrices which allow a factorization FFT where F is some non-negative matrix). The dual of this cone is the cone of copositive matrices (i.e., those matrices which yield a non-negative quadratic form on the positive orthant). This conic formulation allows us to employ primal-dual affine-scaling directions. Furthermore, these approaches are combined with an evolutionary dynamics algorithm which generates primal-feasible paths along which the objective is monotonically improved until a local solution is reached. In particular, the primal-dual affine scaling directions are used to escape from local maxima encountered during the evolutionary dynamics phase.

Completely positive matrices associated with M-matrices

Abstract: (1994). Completely positive matrices associated with M-matrices. Linear and Multilinear Algebra: Vol. 37, No. 4, pp. 303-310.

Nonnegative Matrix Approximation: Algorithms and Applications

Abstract: Low dimensional data representations are crucial to numerous applicatio ns in machine learning, statistics, and signal processing. Nonnegative matrix approximation (NNMA) is a method for dimensionality reduction that respects the nonnegativity of the input data while constructin g a low-dimensional approximation. NNMA has been used in a multitude of applications, though without com mensurate theoretical development. In this report we describe generic methods for minimizing g e eralized divergences between the input and its low rank approximant. Some of our general methods are even extensible to arbitrary convex penalties. Our methods yield efficient multiplicative iterative schemes for s olving the proposed problems. We also consider interesting extensions such as the use of penalty function s, non-linear relationships via “link” functions, weighted errors, and multi-factor approximations. We p r sent some experiments as an illustration of our algorithms. For completeness, the report also includes a brief literature survey of the various algorithms and the applications of NNMA.

Open problems in the theory of completely positive and copositive matrices

Abstract: We describe the main open problems which are currently of interest in the theory of copositive and completely positive matrices. We give motivation as to why these questions are relevant and provide a brief description of the state of the art in each open problem.

Nonnegative matrix factorization : complexity, algorithms and applications

Abstract: Linear dimensionality reduction techniques such as principal component analysis are powerful tools for the analysis of high-dimensional data. In this thesis, we explore a closely related problem, namely nonnegative matrix factorization (NMF), a low-rank matrix approximation problem with nonnegativity constraints. More precisely, we seek to approximate a given nonnegative matrix with the product of two low-rank nonnegative matrices. These nonnegative factors can be interpreted in the same way as the data, e.g., as images (described by pixel intensities) or texts (represented by vectors of word counts), and lead to an additive and sparse representation. However, they render the problem much more difficult to solve (i.e., NP-hard). A first goal of this thesis is to study theoretical issues related to NMF. In particular, we make connections with well-known problems in graph theory, combinatorial optimization and computational geometry. We also study computational complexity issues and show, using reductions from the maximum-edge biclique problem, NP-hardness of several low-rank matrix approximation problems, including the rank-one subproblems arising in NMF, a problem involving underapproximation constraints (NMU) and the unconstrained version of the factorization problem where some data is missing or unknown. Our second goal is to improve existing techniques and develop new tools for the analysis of nonnegative data. We propose accelerated variants of several NMF algorithms based on a careful analysis of their computational cost. We also introduce a multilevel approach to speed up their initial convergence. Finally, a new greedy heuristic based on NMU is presented and used for the analysis of hyperspectral images, in which each pixel is measured along hundreds of wavelengths, which allows for example spectroscopy of satellite images.