Hormann W.,Boazii University |
Sak H.,Vienna University of Economics and Business |
Sak H.,Istanbul Kultur University
Mathematics and Computers in Simulation | Year: 2010
The standard method for generating multi-t vectors is simple and convenient but it has the disadvantage that the generated multi-normal and multi-t vectors are not similar. For t-copula models this destroys much of the variance reduction when using the result of the multi-normal model as external control variate. Therefore we develop a new generation method for multi-t vectors. It is based on the polar method and numerical inversion, and generates multi-normal and multi-t vectors that are very similar. Numerical experiments with simple functions of the weighted sum of t-copula vectors and with pricing European basket options with a t-copula model confirm that the obtained variance reduction factors of the new method are high; 2-100 times higher than when using the standard generation method. © 2010 IMACS.
Gonen M.,Boazii University |
Alpaydn E.,Boazii University
Pattern Recognition | Year: 2011
In recent years, several methods have been proposed to combine multiple kernels using a weighted linear sum of kernels. These different kernels may be using information coming from multiple sources or may correspond to using different notions of similarity on the same source. We note that such methods, in addition to the usual ones of the canonical support vector machine formulation, introduce new regularization parameters that affect the solution quality and, in this work, we propose to optimize them using response surface methodology on cross-validation data. On several bioinformatics and digit recognition benchmark data sets, we compare multiple kernel learning and our proposed regularized variant in terms of accuracy, support vector count, and the number of kernels selected. We see that our proposed variant achieves statistically similar or higher accuracy results by using fewer kernel functions and/or support vectors through suitable regularization; it also allows better knowledge extraction because unnecessary kernels are pruned and the favored kernels reflect the properties of the problem at hand. © 2010 Elsevier Ltd. All rights reserved.
Toku A.E.,Boazii University |
Tekir S.D.,Boazii University |
Ozbayraktar F.B.K.,Boazii University |
Ulgen K.O.,Boazii University
Computational Biology and Chemistry | Year: 2011
In the last few years, researchers have an intense interest in the evolutionarily conserved signaling pathways which have crucial roles during embryonic development. The most intriguing factor of this interest is that malfunctioning of these signaling pathways (Hedgehog, Notch, Wnt etc.) leads to several human diseases, especially to cancer. This study deals with the β-catenin dependent branch of Wnt signaling and the Hedgehog signaling pathways which offer potential targeting points for cancer drug development. The identification of all proteins functioning in these signaling networks is crucial for the efforts of preventing tumor formation. Here, through integration of protein-protein interaction data and Gene Ontology annotations, Wnt/β-catenin and Hedgehog signaling networks consisting of proteins that have statistically high probability of being biologically related to these signaling pathways were reconstructed in Drosophila melanogaster. Next, by the structural network analyses, the crucial components functioning in these pathways were identified. The proteins Arm, Frizzled receptors (Fz and Fz2), Arr, Apc, Axn, Ci and Ptc were detected as the key proteins in these networks. Futhermore, the hub protein Mer having tumor suppressor function may be proposed as a putative drug target for cancer and deserves further investigation via experimental methods. Finally, the crosstalk analysis between the reconstructed networks reveals that these two signaling networks crosstalk to each other. © 2011 Elsevier Ltd. All rights reserved.