Catálogo de publicaciones - libros

One problem of classical phase shift technique for 3D surface measurement is the occurrence of ambiguities due to the use of fringe projection. We propose a universal theory to calculate unambiguous values called projector coordinates. The projector coordinates can be used as a base for a reliable surface reconstruction without any ambiguity. The essence of our method is the application of pattern sequences with different periods. In contrast to combined techniques like hierarchical phase shift or phase shift with Gray code we use all pictures homogeneously which were taken for the measurement. This leads to a higher accuracy. Furthermore we are able to avoid some typical calculation errors that are produced in classical phase shifting.

In a stationary optimization problem, the fitness landscape does not change during the optimization process; and the goal of an optimization algorithm is to locate a stationary optimum. On the other hand, most of the real world problems are dynamic, and stochastically change over time. Genetic Algorithms have been applied to dynamic problems, recently. In this study, we present two hybrid techniques that are applied on moving peaks benchmark problem, where these techniques are the extensions of the leading methods in the literature. Based on the experimental study, it was observed that the hybrid methods outperform the related work with respect to quality of solutions for various parameters of the given benchmark problem.

To provide satisfactory accuracy and flexibility, most of the existing shape retrieval methods make use of different alignments and translations of the objects that introduce much computational complexity. The most computationally expensive part of these algorithms is measuring the degree of match (or mismatch) of the query object with the objects stored in database. In this paper, we present an approach to cut down a large portion of this search space (number of objects in database) that retrieval algorithms need to take into account. This method is applicable in clustering based approaches also. Moreover, this minimization is done keeping the accuracy of the retrieval algorithms intact and its efficiency is not severely affected in high dimensionalities.

Packing problems deal with loading of a set of items (objects) into a set of boxes (containers) in order to optimize a performance criterion under various constraints. With the advance of RFID technologies and investments in IT infrastructures companies now have access to the necessary data that can be utilized in cost reduction of packing processes. Therefore bin packing and container loading problems are becoming more popular in recent years. In this research we propose a beam search algorithm to solve a packing problem that we encountered in a real world project. The 3D-MBSBPP (Multiple Bin Sized Bin Packing Problem) that we present and solve has not been analyzed in literature before, to the best of our knowledge. We present the performance of our proposed beam search algorithm in terms of both cost and computational time in comparison to a greedy algorithm and a tree search enumeration algorithm.

Full search (FS) motion estimation has have been a big problem in real-time video coding because of enormous amount for calculation. Especially, the recent MPEG-4 AVC (advanced video coding) standard requires much more computations in motion estimation than the conventional MPEG-2 coding standard. In this paper, we propose a fast FS algorithm which can reduce only unnecessary computations significantly. Our algorithm is based on selective matching scan and elimination of unlike candidate blocks from initial matching error. According to the obtained matching order, we proceed to calculate matching errors and remove unlike candidate vectors based on PDE (Partial Distortion Elimination) method. Our algorithm takes about 4~10% of computations for block matching error compared with conventional FS algorithm without any degradation in prediction quality, thus our algorithm will be useful to real-time video coding applications using MPEG-4 AVC or MPEG-2 video coding standards.

The Orienteering Problem (OP) is a version of TSP with profits in which instead of a cycle, a path is sought. In this paper, we consider three variations of Variable Neighborhood Search (VNS) and present the first algorithm solely based on VNS to solve the OP. The experimental results for the benchmark problems indicate that the algorithm, designed by using Reduced VNS instead of the local search phase of the traditional VNS, is the best amongst other variations of VNS we tried; it is the most robust and produces the best results, in terms of solution quality, within a reasonable amount of time. Moreover, it improves the best known results for several benchmark problems and reproduces the best results for others.

Finding gene regulation information from microarray time-series data is important to uncover transcriptional regulatory networks. Pearson correlation is the widely used method to find similarity between time-series data. However, correlation approach fails to identify gene regulations if time-series expressions do not have global similarity, which is mostly the case. Assuming that gene regulation time-series data exhibits temporal patterns other than global similarities, one can model these temporal patterns. Hidden Markov models (HMMs) are well established structures to learn and model temporal patterns. In this study, we propose a new method to identify regulation relationships from microarray time-series data using HMMs. We showed that the proposed HMM based approach detects gene regulations, which are not captured by correlation methods. We also compared our method with recently proposed gene regulation detection approaches including edge detection, event method and dominant spectral component analysis. Results on Spellman’s α -synchronized yeast cell-cycle data clearly present that HMM approach is superior to previous methods.

This paper studies the asymptotical lower limits on the required number of samples for identifying Boolean Networks, which is given as Ω( logn ) in the literature for fully random samples. It has also been found that O ( logn ) samples are sufficient with high probability. Our main motivation is to provide tight lower asymptotical limits for samples obtained from time series experiments. Using the results from the literature on random boolean networks, lower limits on the required number of samples from time series experiments for various cases are analytically derived using information theoretic approach.

Computational approach to predict peptide binding to major histocompatibility complex (MHC) is crucial for vaccine design since these peptides can act as a T-Cell epitope to trigger immune response. There are two main branches for peptide prediction methods; structural and data mining approaches. These methods can be successfully used for prediction of T-Cell epitopes in cancer, allergy and infectious diseases. In this paper , association rule mining methods are implemented to generate rules of peptide selection by MHCs. To capture the binding characteristics, modified rule mining and data transformation methods are implemented in this paper. Peptides are known to bind to the same MHC show sequence variability, to capture this characteristic, we used a reduced amino acid alphabet by clustering amino acids according to their physico-chemical properties. Using the classification of amino acids and the OR-operator to combine the rules to reflect that different amino acid types and positions along the peptide may be responsible for binding are the innovations of the method presented. We can predict MHC Class-I binding with 75-97% coverage and 76-100% accuracy.

Functional identification of G-Protein Coupled Receptors (GPCRs) is one of the current focus areas of pharmaceutical research. Although thousands of GPCR sequences are known, many of them are orphan sequences (the activating ligand is unknown). Therefore, classification methods for automated characterization of orphan GPCRs are imperative. In this study, for predicting Level 1 subfamilies of GPCRs, a novel method for obtaining class specific features, based on the existence of activating ligand specific patterns, has been developed and utilized for a majority voting classification. Exploiting the fact that there is a non-promiscuous relationship between the specific binding of GPCRs into their ligands and their functional classification, our method classifies Level 1 subfamilies of GPCRs with a high predictive accuracy between 99% and 87% in a three-fold cross validation test. The method also tells us which motifs are significant for class determination which has important design implications. The presented machine learning approach, bridges the gulf between the excess amount of GPCR sequence data and their poor functional characterization.