Results

**1 - 2**of**2**### Statistics c © Copyright by

, 2009

"... This thesis proposes a methodology for the simultaneous optimization of multiple goal functions via computer experiments. Some technical challenges associated with the black box multiobjective problem (MOP) can be enumerated as follows: the presence of conflicting goals imply that more optimization ..."

Abstract
- Add to MetaCart

This thesis proposes a methodology for the simultaneous optimization of multiple goal functions via computer experiments. Some technical challenges associated with the black box multiobjective problem (MOP) can be enumerated as follows: the presence of conflicting goals imply that more optimization effort is invested to find a good range of solutions that are simul-taneously optimal against these competing criteria; the highly non-linear mapping between the inputs in the design space and the goal functions in objective space may complicate the solution process; and in common with global optimization, the run-time costs of simulation severely limit the number of evaluations that can be made. In view of these, the aim is to compute efficiently and identify a set of good solutions that collectively provide an even coverage of the Pareto front, the set of optimal solutions for a given MOP. The members of the Pareto front comprise the set of compromise solutions from which a decision maker chooses a final design that

### Distance-Based Network Recovery under Feature Correlation

"... We present an inference method for Gaussian graphical models when only pair-wise distances of n objects are observed. Formally, this is a problem of esti-mating an n × n covariance matrix from the Mahalanobis distances dMH(xi,xj), where object xi lives in a latent feature space. We solve the problem ..."

Abstract
- Add to MetaCart

(Show Context)
We present an inference method for Gaussian graphical models when only pair-wise distances of n objects are observed. Formally, this is a problem of esti-mating an n × n covariance matrix from the Mahalanobis distances dMH(xi,xj), where object xi lives in a latent feature space. We solve the problem in fully Bayesian fashion by integrating over the Matrix-Normal likelihood and a Matrix-Gamma prior; the resulting Matrix-T posterior enables network recovery even under strongly correlated features. Hereby, we generalize TiWnet [19], which as-sumes Euclidean distances with strict feature independence. In spite of the greatly increased flexibility, our model neither loses statistical power nor entails more computational cost. We argue that the extension is highly relevant as it yields significantly better results in both synthetic and real-world experiments, which is successfully demonstrated for a network of biological pathways in cancer patients. 1