This work was supported by project PGC2018-098813-B-C31 (Spanish "Ministerio de Ciencia, Innovacion y Universidades") , and by European Regional Development Funds (ERDF) . Funding for open access charge: Universidad de Granada/CBUA. We would like to thank Dr. Daniel Castillo-Secilla and Dr. John Q. Gan for making available to us, respectively, the lung cancer dataset and the University of Essex BCI data files.

Diagnosis of learning difficulties is a challenging goal. There are huge number of factors involved in the evaluation procedure that present high variance among the population with the same difficulty. Diagnosis is usually performed by scoring subjects according to results obtained in different neuropsychological (performance-based) tests specifically designed to this end. One of the most frequent disorders is developmental dyslexia (DD), a specific difficulty in the acquisition of reading skills not related to mental age or inadequate schooling. Its prevalence is estimated between 5% and 12% of the population. Traditional tests for DD diagnosis aim to measure different behavioral variables involved in the reading process. In this paper, we propose a diagnostic method not based on behavioral variables but on involuntary neurophysiological responses to different auditory stimuli. The experiments performed use electroencephalography (EEG) signals to analyze the temporal behavior and the spectral content of the signal acquired from each electrode to extract relevant (temporal and spectral) features. Moreover, the relationship of the features extracted among electrodes allows to infer a connectivity-like model showing brain areas that process auditory stimuli in a synchronized way. Then an anomaly detection system based on the reconstruction residuals of an autoencoder using these features has been proposed. Hence, classification is performed by the proposed system based on the differences in the resulting connectivity models that have demonstrated to be a useful tool for differential diagnosis of DD as well as a method to step towards gaining a better knowledge of the brain processes involved in DD. The results corroborate that nonspeech stimulus modulated at specific frequencies related to the sampling processes developed in the brain to capture rhymes, syllables and phonemes produces effects in specific frequency bands that differentiate between controls and DD subjects. The proposed method showed relatively high sensitivity above 0.6, and up to 0.9 in some of the experiments.

Summary By means of the availability of mechanisms such as Dynamic Voltage and Frequency Scaling (DVFS) and heterogeneous architectures including processors with different power consumption profiles, it is possible to devise scheduling algorithms that are aware of both runtime and energy consumption in parallel programs. In this paper, we propose and evaluate a multi‐objective (more specifically, a bi‐objective) approach to distribute the workload among the processing cores in a given heterogeneous parallel CPU‐GPU architecture. The aim of this distribution may be either to save energy without increasing the running time or to reach a trade‐off among time and energy consumption. The parallel programs considered here are master‐worker evolutionary algorithms where the evaluation of the fitness function for the individuals in the population demands the most part of the computing time. As many useful bioinformatics and data mining applications exhibit this kind of parallel profile, the proposed energy‐aware approach for workload scheduling could be frequently applied.

Summary Bioinformatics applications frequently involve high‐dimensional model building or classification problems that require reducing dimensionality to improve learning accuracy while irrelevant inputs are removed. Thus, feature selection has become an important issue on these applications. Moreover, several approaches for supervised and unsupervised feature selections as a multi‐objective optimization problem have been recently proposed to cope with issues on performance evaluation of classifiers and models. As parallel processing constitutes an important tool to reach efficient approaches that make it possible to tackle complex problems within reasonable computing times, in this paper, alternatives for the cooperation of subpopulations in multi‐objective evolutionary algorithms have been identified and classified, and several procedures have been implemented and evaluated on some synthetic and Brain–Computer Interface datasets. The results show different improvements achieved in the solution quality and speedups, depending on the cooperation alternative and dataset. We show alternatives that even provide superlinear speedups with only small reductions in the solution quality, besides another cooperation alternative that improves the quality of the solutions with speedups similar to, or only slightly higher than, the speedup obtained by the parallel fitness evaluation in a master‐worker implementation (the alternative used as reference that behaves as the corresponding sequential multi‐objective approach). Copyright © 2015 John Wiley & Sons, Ltd.

Feature selection is required in many applications that involve high dimensional model building or classification problems. Many bioinformatics applications belong to this type. Recently, some approaches for supervised and unsupervised feature selection as a multi-objective optimization problem have been proposed. As the performance of unsupervised classification is evaluated through the quality of the obtained groups or clusters in the data set to be classified, it is difficult to define a suitable objective function that drives the selection of the features. Thus, several evaluation measures, and thus multi-objective clustering characterization, could provide a suitable set of features for unsupervised classification. In this paper, we consider the parallel implementation of a multi-objective feature selection that makes it possible to apply it to complex classification problems such as those having many features to select, and specifically high-dimensional data sets with much more features than data items. In this paper, we propose master-worker implementations of two different parallel evolutionary models, the parallel computation of the cost functions for the individuals in the population, and the parallel execution of evolutionary multi-objective procedures on subpopulations. The experiments accomplished on different benchmarks, including some related with feature selection in classification of EEG (Electroencephalogram) signals for BCI (Brain Computer Interface) applications, show the benefits of parallel processing not only for decreasing the running time, but also for improving the solution quality.

Present trend towards multiprocessor and multi-core architectures as well as programmable NICs (Network Interface Cards) provides new opportunities to exploit the available parallelism in the network interface design and implementation to cope with the high communication overhead required to take advantage of a multi-gigabit links. In this paper we have used dynamic web servers to evaluate several alternatives to distribute the network interface work among the different cores available in the node. These alternatives have been devised according to the affinity of the work to be done with the location (proximity to the memories where the different data structures are stored) and characteristics of the processing core. Improvements in the throughput (up to 110%), the response time (up to 500%) and the requests attended per second (up to 85%) have been observed in our experiments on dynamic web servers.

OpenFlow switching enables flexible management of enterprise network switches and experiments on regular network traffic. We present in this paper a complementary design to OpenFlow's existing reference designs. We apply network processor based acceleration cards to perform OpenFlow switching. We describe the design options and report our experiment results that show a 20% reduction on packet delay and the comparable packet forwarding throughput compared to conventional designs.

In this paper, we propose a network interface that takes advantage of the parallelism available in network processors to improve the communication performance. Morespecifically, our proposal is based on the Intel IXP28xx network processor, as it includes sixteen processing cores with multithreading and an optimized design for packet processing. Thus, the proposed interface facilitates the exploitation of different options to optimize the communication path in the host by using network interface offloading and/or onloading strategies that try to take advantage of multi-core processors, which are available even in the network cards. The experimental results obtained for TCP/IP and raw Ethernet packets show that multithreading, implemented through the processing cores of the network processor, improves communication performance in both latency of the packets and bandwidth, thus allowing the applications that present high communication requirements to fulfill them through the use of multithreading.

The availability of multi-core processors and programmable NICs (Network Interface Cards), such as TOEs (TCP/IP Offloading Engines), provides new opportunities for designing efficient network interfaces to cope with the gap between the improvement rates of link bandwidths and microprocessor performance. This gap poses important challenges related with the high computational requirements associated to the traffic volumes and wider functionality that the network interface has to support. An opportunity to reach these goals comes from the exploitation of the parallelism in the communication path by distributing the protocol processing work across processors available in the computer, i.e. multi-core microprocessors and programmable NICs. Thus, alternatives such as offloading and onloading try to release host CPU cycles by this approach. Nevertheless, whereas onloading uses another general-purpose processor, either included in a multi-core microprocessor or in a symmetric multiprocessor (SMP), offloading takes advantage of processors in programmable network interface cards (NICs). Some experimental results demonstrate that the relative improvement on peak throughput offered by offloading and onloading depends on the rate of application workload to communication overhead, the message sizes, and on the characteristics of system architecture, more specifically the bandwidth of the buses and the way the NIC is connected to the system processor and memory. Thus, in this paper we propose a network interface that takes the advantages of both offloading and onloading approaches while avoids their respective drawbacks. The performance analyses done by using a full-system simulator, shows that, in the benchmarks and application used for the experiments, our hybrid interface improves the latency and bandwidth behavior of the onloading and offloading approaches.

The protein structure prediction (PSP) problem is considered an open problem as there is no recognized "best" procedure to find solutions. Moreover, this problem presents a vast search space and the analysis of each conformation requires significant amount of computing time. Thus, parallel processing could provide new opportunities to improve the quality of the solutions found until now. In this paper we use a parallel multiobjective evolutionary approach to the problem including a method to decrease the complexity of the problem by reducing the set of variables involved in the optimization process. The proposed parallel procedure provides linear speedups in the structure prediction of the proteins used as benchmarks.

This paper compares the onloading and offloading alternatives for improving up communication. Both strategies try to release host CPU cycles by taking advantage of the execution of the communication workload in other processors present in the node. Nevertheless, whereas onloading uses another general-purpose processor, either included in a chip multiprocessor (CMP) or in a symmetric multiprocessor (SMP), offloading takes advantage of processors in programmable network interface cards (NICs). Here, it is shown that the relative improvement on peak throughput offered by offloading and onloading depends on the rate of application workload to communication overhead, the message sizes, and the characteristics of system architecture, more specifically the buses bandwidth and the way the NIC is connected to the system processor and memory. In our implementations, offloading provides lower latencies than onloading although the CPU utilization and interrupts are lower for onloading.

This paper deals with the use of parallel processing for multi-objective optimization in applications in which the objective functions, the restrictions, and hence also the solutions can change over time. These dynamic optimization problems appear in quite different real-world applications with relevant socio-economic impact. The procedure in this paper is presented based on PSFGA, a parallel evolutionary procedure for multi-objective optimization. It uses a master process that distributes the population among the processors in the system (that evolve their corresponding solutions according to an island model), and collects and adjusts the set of local Pareto fronts found by each processor (this way, the master also allows an implicit communication among islands). Moreover, the procedure exclusively uses non-dominated individuals for the selection and variation, and maintains the diversity of the approximation to the Pareto front by using a strategy based on a crowding distance.

Due to the network technology advances, an order-of-magnitude jump has been produced in the network bandwidth. This fact has reawake the interest in protocol offloading (particularly in protocols such as TCP, that require a lot of CPU resources to process the stack), since network communication is one key factor for the system performance. Nevertheless, there are controversial studies on protocol offloading benefits, and much research work on offloading, particularly on TCP offloading as it has been the main transport protocol for many years, and in fact, it is the most used protocol in the networks. This paper describes a model to simulate by Simics the protocol offloading technique to keep low the communication overheads on the host CPU. The simulation platform provided by Simics can be used for a functional system simulation, including the application program, the operating system, the protocol stack and the device drivers, since network-oriented applications require a full-system simulation. As Simics does not provide a detailed network I/O model, in this paper we describe the way we have generated a model that allows the simulation of the offloading effects

This paper describes our work with Simics (Magnusson, 2002) in the evaluation of protocol offloading to keep low the communication overheads on the host CPU. Simics is a platform that can be used for a functional system simulation, including the application program, the operating system, the protocol stack and the device drivers. Nevertheless, network-oriented applications require a full-system simulation that includes not only the OS code, but also a model of the memory system and a detailed timing model of network DMA effects (Binkert, 2003). As Simics does not provide a detailed network I/O model, it could present some limitations that make it difficult an accurate network-oriented full-system simulation. Here we present the way we have overcome this problem

Evolutionary algorithms provide ways to explore wide search spaces. Thus, it is possible to get some conclusions about the characteristics of these spaces in order to aid in the determination of the best alternatives to solve the problem at hand. We have applied a genetic algorithm to assess the problem of distributed load balancing in parallel processing. To do that, we propose a classification of the space of design of distributed load balancing algorithms that takes into account the different alternatives for each dimension of the algorithm. This classification allows the codification of each load balancing strategy, thus making possible to apply a genetic search to determine the distributed load balancing procedure that provides the best performance for the type of parallel application at hand and the parallel platform where it is implemented. As an example, in this paper we provide the results corresponding to the parallel multiplication of matrices implemented in a cluster.

This paper presents a multiobjective evolutionary algorithm to optimize radial basis function neural networks (RBFNNs) in order to approach target functions from a set of input-output pairs. The procedure allows the application of heuristics to improve the solution of the problem at hand by including some new genetic operators in the evolutionary process. These new operators are based on two well-known matrix transformations: singular value decomposition (SVD) and orthogonal least squares (OLS), which have been used to define new mutation operators that produce local or global modifications in the radial basis functions (RBFs) of the networks (the individuals in the population in the evolutionary procedure). After analyzing the efficiency of the different operators, we have shown that the global mutation operators yield an improved procedure to adjust the parameters of the RBFNNs.

This paper proposes the fusion of two important paradigms, Genetic Algorithms and the Blind Separation of Sources in Nonlinear Mixtures (GABSS). Although the topic of BSS, by means of various techniques, including ICA, PCA, and neural networks, has been amply discussed in the literature, the possibility of using genetic algorithms has not been explored thus far. However, in Nonlinear Mixtures, optimization of the system parameters and, especially, the search for invertible functions is very difficult due to the existence of many local minima. From experimental results, this paper demonstrates the possible benefits offered by GAs in combination with BSS, such as robustness against local minima, the parallel search for various solutions, and a high degree of flexibility in the evaluation function.

To date, clustering techniques have always been oriented to solve classification and pattern recognition problems. However, some authors have applied them unchanged to construct initial models for function approximators. Nevertheless, classification and function approximation problems present quite different objectives. Therefore it is necessary to design new clustering algorithms specialized in the problem of function approximation. This paper presents a new clustering technique, specially designed for function. approximation problems, which improves the performance of the approximator system obtained, compared with other models derived from traditional classification oriented clustering algorithms and input-output clustering techniques.

The CLIC is a lightweight protocol that has recently been proposed for efficient communication in clusters using the Linux Operating System. Besides optimizing communication performance, by reducing the latencies and increasing the bandwidth figures even for short messages, the proposed communication layer also meets other requirements such as multiprogramming, portability, protection against corrupted programs, reliable message delivery, direct access to the network for all applications, etc. In this way, instead of removing the operating system kernel from the critical path and creating a user-level network interface, the operating system support has been optimized to provide reliable and efficient network software, avoiding the TCP/IP protocol stack. The LAM-MPI communication layer has been implemented on top of the proposed protocol and the communication performance has been tested in a cluster of PCs with Linux OS and interconnected with Fast Ethernet.

Abstract The parallelization of complex planning and control problems arising in diverse application areas in the industrial, services and commercial environments not only allows the determination of control variables in the required times but also improves the performance of the control procedure as more processors are involved in the execution of the parallel program. In this paper we describe a scheduling application in a water supply network to demonstrate the benefits of parallel processing. The procedure we propose combines dynamic programming with genetic algorithms and time series prediction in order to solve problems in which decisions are made in stages, and the states and control belong to a continuous space. Taking into account the computational complexity of these applications and the time constraints that are usually imposed, the procedure has been implemented by a parallel program in a cluster of computers, an inexpensive and widely extended platform that can make parallelism a practical means of tackling complex problems in many different environments. Copyright © 2001 John Wiley & Sons, Ltd.

An analysis of the influence of weight and input perturbations in a multilayer perceptron (MLP) is made in this article. Quantitative measurements of fault tolerance, noise immunity, and generalization ability are provided. From the expressions obtained, it is possible to justify some previously reported conjectures and experimentally obtained results (e.g., the influence of weight magnitudes, the relation between training with noise and the generalization ability, the relation between fault tolerance and the generalization ability). The measurements introduced here are explicitly related to the mean squared error degradation in the presence of perturbations, thus constituting a selection criterion between different alternatives of weight configurations. Moreover, they allow us to predict the degradation of the learning performance of an MLP when its weights or inputs are deviated from their nominal values and thus, the behavior of a physical implementation can be evaluated before the weights are mapped on it according to its accuracy.

Genetic algorithms, time series prediction, and Monte Carlo simulation are applied to dynamic programming in order to solve complex planning and control problems in which decisions are made in stages, and the states and control belong to a continuous space. Each decision has an immediate associated cost and also affects the cost of future stages. Therefore, a balance is required between a low cost solution at the present and the possible high costs in the future. A hybrid genetic algorithm is used to determine the feasible functioning states in each stage. A procedure for series prediction based on RBF networks allows the uncertainty about state transitions to be avoided and Monte Carlo simulations are used to approximate the cost-to-go function, thus reducing the computational cost of the dynamic programming procedure. As an example, the proposed procedure is applied to a water supply network scheduling problem.

This paper describes a procedure for controlling the water supply system. The controller uses a neural network to predict the water demand levels and a genetic algorithm to determine the feasible operation points in an optimal strategy that is based on dynamic programming. The controller has been executed in parallel in a cluster of computers. This has allowed not only the determination of the control commands in the required times but also the improvement of the control procedure performances.

Describes a structure to create a RBF neural network. This structure has 4 main characteristics. The first one is that the special RBF network architecture uses regression weights to replace the constant weights normally used. These regression weights are assumed to be functions of input variables. The second characteristic is the normalization of the activation of the hidden neurons (weighted average) before aggregating the activations, which, as observed by various authors, produces better results than the classical weighted sum architecture. The third aspect is that a new type of nonlinear function is proposed, the pseudo-gaussian function (PGBF). With this, the neural system gains flexibility, as the neurons possess an activation field that does not necessarily have to be symmetric with respect to the centre or to the location of the neuron in the input space. In addition to this new structure, we propose, as the fourth and final feature, a sequential learning algorithm, which is able to adapt the structure of the network, with this, it is possible to create new hidden units and also to detect and remove inactive units.

This paper presents three parallel procedures implemented in a shared-memory multiprocessor to generate the patterns that allow the testing of digital circuits. The implementation of these procedures in a multiprocessor uses the system memory better than in a distributed-memory multicomputer, since it is not necessary to store the circuit structure in the local memory of each processor, besides other common structures. The parallel test generation procedures are based on a new sequential algorithm which mixes both the Boolean difference and digital spectral techniques. It is thus different from other methods proposed that deal with the parallelization of test generation algorithms that carry out an implicit enumeration of the input pattern space. The first procedure distributes the set of faults using a backtracking procedure starting from a primary output and allocating a similar number of lines to each processor. The second procedure distributes the set of faults among the processors taking into account the distance from each line to its nearest primary output; it then applies the algorithm to generate the test pattern with some modifications. The third procedure uses a circuit partitioning procedure which allows similar sized parts of the circuit to be assigned to each processor while communications between processors are minimized. The experimental results obtained when the procedures are applied to the usual benchmark circuits (the ISCAS set) show figures for speedup better than in a multicomputer, although fewer processors are used. Copyright © 2000 John Wiley & Sons, Ltd.

In the synthesis of a fuzzy system two steps are generally employed: the identification of a structure and the optimization of the parameters defining it. The paper presents a methodology to automatically perform these two steps in conjunction using a three-phase approach to construct a fuzzy system from numerical data. Phase 1 outlines the membership functions and system rules for a specific structure, starting from a very simple initial topology. Phase 2 decides a new and more suitable topology with the information received from the previous step; it determines for which variable the number of fuzzy sets used to discretize the domain must be increased and where these new fuzzy sets should be located. This, in turn, decides in a dynamic way in which part of the input space the number of fuzzy rules should be increased. Phase 3 selects from the different structures obtained to construct a fuzzy system the one providing the best compromise between the accuracy of the approximation and the complexity of the rule set. The accuracy and complexity of the fuzzy system derived by the proposed self-organized fuzzy rule generation procedure (SOFRG) are studied for the problem of function approximation. Simulation results are compared with other methodologies such as artificial neural networks, neuro-fuzzy systems, and genetic algorithms.

In this paper, a systematic design is proposed to determine fuzzy system structure and learning its parameters, from a set of given training examples. In particular, two fundamental problems concerning fuzzy system modeling are addressed: 1) fuzzy rule parameter optimization and 2) the identification of system structure (i.e., the number of membership functions and fuzzy rules). A four-step approach to build a fuzzy system automatically is presented: Step 1 directly obtains the optimum fuzzy rules for a given membership function configuration. Step 2 optimizes the allocation of the membership functions and the conclusion of the rules, in order to achieve a better approximation. Step 3 determines a new and more suitable topology with the information derived from the approximation error distribution; it decides which variables should increase the number of membership functions. Finally, Step 4 determines which structure should be selected to approximate the function, from the possible configurations provided by the algorithm in the three previous steps. The results of applying this method to the problem of function approximation are presented and then compared with other methodologies proposed in the bibliography.

An evolutionary computation approach is used to learn online the rules that allow the processors in a parallel platform to cooperate by interchanging the local optima that they find while they concurrently explore different zones of the solution space. The cooperation of processors can greatly benefit the resolution of combinatorial optimization problems by decreasing their runtimes, by increasing the quality of the solutions obtained, or both. Moreover, as parallel computers are more and more accessible, the application of parallel processing to solve these problems becomes a practical and interesting alternative. As an example, a parallel optimization algorithm based on Boltzmann Machine has been used for a detailed description and evaluation of the proposed cooperation approach.

This paper shows a quantitative relation between the regularization techniques, the generalization ability, and the sensitivity of the Multilayer Perceptron (MLP) to input noise. Although many studies about these topics have been presented, in most cases only one of the problems is addressed, and only experimentally obtained evidence is provided to illustrate some kind of correlation between generalization, noise immunity and the use of regularization techniques to obtain a set of weights after training that provides the corresponding MLP with generalization ability and noise immunity. Here, a new measurement of noise immunity for a MLP is presented. This measurement, which is termed Mean Squared Sensitivity (MSS), explicitly evaluates the Mean Squared Error (MSE) degradation of a MLP when it is perturbed by input noise, and can be computed from the statistical sensitivities (previously proposed) of the output neurons. The MSS provides an accurate evaluation of the MLP performance loss when its inputs are perturbed by noise and can also be considered a measurement of the smoothness of the error surface with respect to the inputs. Thus, as the MSS can be used to evaluate the noise immunity or the generalization ability, it gives a criterion to select among different weight configurations that present a similar MSE after training.

The use of generalized Hopfield neural networks in designing the checking circuitry of a concurrent testable circuit is discussed. The aliasing probability, a measure for evaluating the performance of the checking circuitry, is provided. It is shown how, by using spectral techniques based on the Reed-Muller transform, the aliasing probability can be expressed as a function of the Reed-Muller coefficients. Therefore, obtaining the checking circuitry means selecting a set of Reed-Muller spectral coefficients, with fewer elements than a given bound, that minimizes the aliasing probability. To apply the neural networks to design the checking circuitry for concurrent testing, the aliasing probability has been used as an energy function, and the Hopfield neural network has been modified to have an associated energy function with any type of polynomial dependence on the processor states.>

Reed-Muller coefficients are used to generate a test-pattern selection procedure for detecting single stuck-at faults. This procedure is based on the heuristics deduced from the way in which the spectral coefficients are affected by such faults. The changes that the spectral coefficients undergo are also compared with the fault models used most frequently to model defects in combinational circuits. The proposed pattern-generating procedure does not need a simulation of the circuit for each of the possible stuck-at faults, and its complexity is proportional to the number of gates in the circuit. The proportionality constant increases exponentially as the number of inputs in the circuit increases. To evaluate the performance of the proposed method, its application to some benchmark circuits, including the ALU 74181, is presented.>

Several problems in switching theory such as automatic test pattern generation, can be exactly and elegantly investigated by using switching equations. An efficient tabular method is presented to solve these switching equations. The solutions of a given equation are compacted into a table, and a Boolean algebra is defined for these tables. The proposed procedure is based on the use of this Boolean algebra. The compactness and simplicity of the method are outstanding, and it is straightforward to implement for computer processing. The complexity of the procedure is computed, and some experimental results for a set of benchmark equations are provided in order to point out the effectiveness of this method.>