Kwoh Chee Keong

Extreme learning machine for predicting HLA-Peptide binding

Machine learning techniques have been recognized as powerful tools for learning from data. One of the most popular learning techniques, the Back-Propagation (BP) Artificial Neural Networks, can be used as a computer model to predict peptides binding to the Human Leukocyte Antigens (HLA). The major advantage of computational screening is that it reduces the number of wet-lab experiments that need to be performed, significantly reducing the cost and time. A recently developed method, Extreme Learning Machine (ELM), which has superior properties over BP has been investigated to accomplish such tasks. In our work, we found that the ELM is as good as, if not better than, the BP in term of time complexity, accuracy deviations across experiments, and – most importantly – prevention from over-fitting for prediction of peptide binding to HLA.

Parallel DNA sequence alignment on the cell broadband engine

Sequence alignment is one of the most important techniques in Bioinformatics. Although efficient dynamic programming algorithms exist for this problem, the alignment of very long DNA sequences still requires significant time on traditional computer architectures. In this paper, we present a scalable and efficient mapping of DNA sequence alignment onto the Cell BE multicore architecture. Our mapping uses two types of parallelization techniques: (i) SIMD vectorization within a processor and (ii) wavefront parallelization between processors.

Fast leave-one-out evaluation and improvement on inference for LS-SVMs

In this paper, a fast leave-one-out (LOO) evaluation formula is introduced for least squares support vector machine (LS-SVM) classifiers. The computation cost can be reduced to approximately 1/N when compared to normal LOO procedure (N is the number of training samples). Inspired by its fast speed, we are able to use it to replace the original level 3 posterior probability approximation formula of the Bayesian framework for LS-SVM classifiers. The improved inference framework shows higher generalization performance and faster computation speed.

Using classification for constrained memetic algorithm: A new paradigm

Regression has been successfully combined with the memetic algorithm (MA) for constructing surrogate models. It is essentially an attempt to approximate the objective or constraint landscape of a constrained optimization problem. Classification, on the other hand, has probably never been thought of being of any assistance to the MA. In fact, it can be used to approximate the feasibility boundary by means of some decision functions. The search effort can thus be focussed on the nearby region, recalling that many constrained optimization problems have their optimal solutions situated on the boundaries. This simply means that only potential individuals will undergo local refinements, reducing the number of function evaluations and accelerating the identification of the global optimum. Presented in this paper is a new approach that combines the support vector machine (SVM) with the MA to achieve this purpose.

Identifying simple discriminatory gene vectors with an information theory approach

In the feature selection of cancer classification problems, many existing methods consider genes individually by choosing the top genes which have the most significant signal-to-noise statistic or correlation coefficient. However the information of the class distinction provided by such genes may overlap intensively, since their gene expression patterns are similar The redundancy of including many genes with similar gene expression patterns results in highly complex classifiers. According to the principle of Occams razor, simple models are preferable to complex ones, if they can produce comparable prediction performances to the complex ones. In this paper, we introduce a new method to learn accurate and low-complexity classifiers from gene expression profiles. In our method, we use mutual information to measure the relation between a set of genes, called gene vectors, and the class attribute of the samples. The gene vectors are in higher-dimensional spaces than individual genes, therefore, they are more diverse, or contain more information than individual genes. Hence, gene vectors are more preferable to individual genes in describing the class distinctions between samples since they contain more information about the class attribute. We validate our method on 3 gene expression profiles. By comparing our results with those from literature and other well-known classification methods, our method demonstrated better or comparable prediction performances to the existing methods, however, with lower-complexity models than existing methods.

3D Posture Reconstruction and Human Animation from 2D Feature Points

An optimal approach is proposed in this paper for posture reconstruction and human animation from 2D feature points extracted from the monocular images containing human motions. Biomechanical constraints are encoded in every joint of the adopted 3D skeletal human model to make sure that each state of the joints represents a physically valid posture. Size of the human model is adjusted to be consistent with the human figure represented by feature points. Energy Function is defined to represent the residuals between the extracted 2D feature points and the corresponding features resulted from projection of the 3D human model. Local Adjustment and Global Adjustment procedures are proposed to place the joints and body segments into proper locations and orientations in 3D space to create the posture with the minimum value of Energy Function. To find the optimal solution of the ill‐posed recovery problem from 2D to 3D, Genetic Algorithm is employed in the high‐dimensional parameter space by considering all the parameters simultaneously. Smooth and continuous changes between consecutive frames are considered in development of the human animation procedure. The proposed approach produces optimal reconstruction results of any possible human postures and movements. It is different from classical kinematics and dynamics formulations, and is an attempt to bridge the gap between computer vision and computer animation in human motion study.

Classification-assisted memetic algorithms for solving optimization problems with restricted equality constraint function mapping

The success of Memetic Algorithms (MAs) has driven many researchers to be more focused on the efficiency aspect of the algorithms such that it would be possible to effectively employ MAs to solve computationally expensive optimization problems where single evaluation of the objective and constraint functions may require minutes to hours of CPU time. One of the important design issues in MAs is the choice of the individuals upon which local search procedure should be applied. Selecting only some potential individuals lessens the demand for functional evaluations hence accelerates convergence to the global optimum. In recent years, advances have been made targeting optimization problems with single equality constraint h(x) = 0. The presence of previously evaluated candidate solutions with different signs of constraint values within some localities thus allows the estimation of the constraint boundary. An individual will undergo local search only if it is sufficiently close to the approximated boundary. Elegant as it may seem, the approach had unfortunately assumed that every constraint function maps the design variables to optimize into unbounded real values. This, however, may not always be the case in practice. In this paper, we present a strategy to efficiently solve constrained problems with a single equality constraint; the function of which maps the design variables into restricted (either strictly non-negative or strictly non-positive) real values only.

Applications of Heterogeneous Structure of Cell Broadband Engine Architecture for Biological Database Similarity Search

In this paper, we present a technique to optimize the performance of database similarity search in the specific context of Cell Broadband Engine Architecture (CBEA). The technique applied was Striped Smith-Waterman algorithm for SIMD and heterogeneous task distribution in MIMD.In terms of sensitivity, the technique preserves the optimality of original Smith- Waterman algorithm. In addition, the performance recorded shows a remarkable speedup of 1.7 to 8.8 folds of this new architecture, as compared to other platforms such as Streaming SIMD Extensions 2 (SSE2) and Graphics Processing Unit (GPU).

MRI head segmentation for object based volume visualization

In this paper, we present a new image segmentation approach for MRI of the head, which is a semi-automatic process. Unlike automatic segmentation or manual segmentation, the semi-automatic segmentation approach is a robust and interactive segmentation process. This approach carries out 3D volume data segmentation based on 2D image slices. By utilising the user-provided image mask, including areas of interest or structural information, the semi-automatic segmentation process can generate a new segmented volume dataset and structural information. The object based volume visualization method can use this segmented dataset and structural information to perform structure based manipulation and visualization, which cannot be achieved using a normal volume rendering method.

MOTION RECOVERY BASED ON FEATURE EXTRACTION FROM 2D IMAGES

This paper presents a method for motion recovery from monocular images containing human motions. Image processing techniques, such as spatial filter, linear prediction, cross correlation, least square matching etc, are applied to extract feature points from 2D human figures with or without markers. A 3D skeleton human model is adopted with encoded angular constraints. Energy Function is defined to represent the residuals between extracted feature points and the corresponding points resulted from projecting the human model to the projection plane. Then a procedure for motion recovery is developed, which makes it feasible to generate realistic human animations