Andrés Bueno-Crespo

Soft Computing Techniques for the Protein Folding Problem on High Performance Computing Architectures

The protein-folding problem has been extensively studied during the last fifty years. The understanding of the dynamics of global shape of a protein and the influence on its biological function can help us to discover new and more effective drugs to deal with diseases of pharmacological relevance. Different computational approaches have been developed by different researchers in order to foresee the threedimensional arrangement of atoms of proteins from their sequences. However, the computational complexity of this problem makes mandatory the search for new models, novel algorithmic strategies and hardware platforms that provide solutions in a reasonable time frame. We present in this revision work the past and last tendencies regarding protein folding simulations from both perspectives; hardware and software. Of particular interest to us are both the use of inexact solutions to this computationally hard problem as well as which hardware platforms have been used for running this kind of Soft Computing techniques.

Deep Learning Approach for Classifying Papanicolau Cervical Smears

Cervical cancer is the third neoplasm in frequency worldwide between women. Screening techniques in general population have demonstrated clear effectiveness as its implementation has decreased cervical cancer incidence and mortality more than 70% in several countries. This benefit is related with detection of early pre-malignant asymptomatic lesions, that can be treated to avoid their progression to invasive cancer. Papanicolau cervical smear is the most common cancer screening technique worldwide used since described by Giorgios Papanicolau on 1928. Sampling techniques have been improved in last years, based on simplifying and automatizing procedures. However, after preparing the samples, an expert review of the microscopic images is needed. There are few automatic diagnostic methods published, but their results are not as good as an expert examination. In this paper, we develop a microscopic cervical cells database using Papanicolau cervical smears from our patients, sampled few minutes before performing a cone biopsy on them. With this procedure, we have both the cervical smear and the biopsy diagnostics, tagged as Gold Standard. Then, a deep-learning methodology is performed for the automatic categorization of pre-malignant and benign cervical cells. We use the the Caffe deep-learning framework to leverage NVIDIA GPU computing architectures to deal with this real patient database in a reduced time-frame. Our results reveal the deep learning methodology is robust in this biomedical classification, reaching up to 78% Accuracy.

Multi-objective optimal design of submerged arches using extreme learning machine and evolutionary algorithms

Abstract The design of funicular (or momentless) submerged arches has a great application in the fields of the building construction and the civil engineering. Traditional approaches in this field have been based on the resolution of ordinary differential equations that govern the structural behavior of the submerged arches. Indeed, these approaches only consider a design parameter and they are computationally expensive. For intermediate depth ratios, the funicular shape of the arch lays about halfway between the geometric forms of the parabola and the ellipse. Actually, the arch centerline could be modeled as a parametric linear function of these two conical shapes where different parameters are established, opening new opportunities for the optimization in the design of such structures, which also consider several design parameters. In this article, we propose a methodology to optimize several parameters in the design of submerged arches. Specifically, we focus on the reduction of the arch bending moment, which is a critical factor in the design cost of the structure, and also the maximization of the airspace enclosed by the arch, which is of particular interest in the serviceability of recreational submerged installations. Our methodology is based on a multi-objective evolutionary algorithm, which uses artificial neural networks with extreme learning machine (ELM) to predict the level of bending stresses at the submerged arch under different shape configurations and also reduce the overall computational cost. Two groups of test examples, corresponding to deep and shallow waters, are developed to compare the numerical results obtained by multi-objective optimization with the theoretical curves predicted by the traditional funicular analysis. Our experimental results offer good accuracy (R2 up to 93%) in the fitness evaluation using ELM. After the multi-objective optimization procedure, our results show optimal arch-shapes with minimum bending stress (i.e., minimum cost) and maximum airspace; thus, the functionality of the underwater installation is also optimal.

Bioinspired Architecture Selection for Multitask Learning

Faced with a new concept to learn, our brain does not work in isolation. It uses all previously learned knowledge. In addition, the brain is able to isolate the knowledge that does not benefit us, and to use what is actually useful. In machine learning, we do not usually benefit from the knowledge of other learned tasks. However, there is a methodology called Multitask Learning (MTL), which is based on the idea that learning a task along with other related tasks produces a transfer of information between them, what can be advantageous for learning the first one. This paper presents a new method to completely design MTL architectures, by including the selection of the most helpful subtasks for the learning of the main task, and the optimal network connections. In this sense, the proposed method realizes a complete design of the MTL schemes. The method is simple and uses the advantages of the Extreme Learning Machine to automatically design a MTL machine, eliminating those factors that hinder, or do not benefit, the learning process of the main task. This architecture is unique and it is obtained without testing/error methodologies that increase the computational complexity. The results obtained over several real problems show the good performances of the designed networks with this method.

Fetal MRI as Complementary Study of Congenital Cystic Adenomatoid Malformation During Pregnancy: A Single Case Report

Fetal lung masses are rare findings in prenatal ultrasound scanning in general population, of which congenital cystic adenomatoid malformation is the most commonly diagnosed type. This paper reports a single case of congenital cystic adenomatoid malformation detected at our hospital and the subsequent clinical follow-up using ultrasound scanning and fetal magnetic resonance imaging.