This paper introduces the inverse modeling constrained multi-objective evolutionary algorithm based on decomposition (IM-C-MOEA/D) for addressing constrained real-world optimization problems. Our research builds upon the advancements made in evolutionary computing-based inverse modeling, and it strategically bridges the gaps in applying inverse models based on decomposition to problem domains with constraints. The proposed approach is experimentally evaluated on diverse real-world problems (RWMOP1-35), showing superior performance to state-of-the-art constrained multi-objective evolutionary algorithms (CMOEAs). The experimental results highlight the robustness of the algorithm and its applicability in real-world constrained optimization scenarios.

Constrained multiobjective optimization problems (CMOPs) are commonly found in real-world applications. CMOP is a complex problem that needs to satisfy a set of equality or inequality constraints. This paper proposes a variant of the bidirectional coevolution algorithm (BiCo) with differential evolution (DE). The novelties in the model include the DE differential mutation and crossover operators as the main search engine and a non-dominated sorting selection scheme. Experimental results on two benchmark test suites and eight real-world CMOPs suggested that the proposed model reached better overall performance than the original model.

We propose a new voice commanded system using speech recognition for robots. This RoboticsysTool System performs navigation instructed by isolated-word guiding linear and angular velocity of the TIAGo robot. We also tested the navigation of the TIAGo robot between points of interest commanded by short sentences. We have chosen the Speech Commands database and the Librispeech database to validate the isolated word recognizer and the continuous speech recognizer. We also added babble noise, using the ADDNoiseTool noise addition tool, to the noise generated by the engine TIAGo robot. We reached promising results with this real-time system. Thus, the execution of the MLP and RNN neural networks in this context of the babble noise obtained a satisfactory response for TIAGo robot navigation tasks.

The inverse modeling multi-objective evolutionary algorithm (IM-MOEA) is a method to solve multi-objective optimization problems (MOP) that samples candidate solutions straight from the objective space, making it easier to control the diversity of the solutions. In the literature, the objective space is partitioned into several subregions by predefining a set of reference vectors, and the selection criterion adopted is based on dominance. These features can cause difficulties to deal with large-scale MOPs (LSMOPs) and with many-objective optimization problems (MaOPs). To address such an issue, this paper proposes the IM-MOEA based on decomposition (IM-MOEA/D) which uses a new scheme for grouping in the objective space based on k-means and a selection criterion based on decomposition, global replacement, that chooses the most appropriate reference vector from the whole population. The experimental results on 45 LSMOPs for 2 to 6 objectives suggest that IM-MOEA/D reached better performance than the compared state-of-the-art MOEAs.

Traditional Combinatorial Reverse Auctions (CRAs) (multiple items and single or multiple attributes) have been effectively adopted in several real-world applications. However, looking for the best solution (a set of winning sellers) is difficult to solve due to CRAs complexity. The use of exact algorithms is quite unsuitable in some real-life auctions due to the exponential time cost of these algorithms. Hence, we have opted for multi-objective evolutionary optimization methods (MOEAs) to find the best compromising solutions. This paper makes a comparative study between different MOEAs to solve the Problem of Determination of Winners (WDP) in a Multi Attribute Combinatorial Reverse Auctions (MACRA) of several items with several attributes, establishing a model inspired by a WDP found in the literature. Six real problems of purchasing electronic products of different complexities are considered. The algorithms were assessed according to MOEA evaluation metrics and according to the best auction solution.

Decomposition-based evolutionary algorithms that work with an appropriate set of weights might obtain a quality final solution set in spite of the use of uniformly distributed and fixed weights that has two important limitations: it may fail depending on the problem geometry; and the population size is not flexible when dealing with Many-objective Problems (MaOPs). Recently proposed, the MOEA/D with Uniformly Randomly Adaptive Weights (MOEA/D-URAW) deals with these limitations using uniformly randomly weights generation method and weight adaptation based on the population sparsity. This paper validates this new approach, the MOEA/D-URAW, with state-of-the-art evolutionary algorithms in MaOPs, i.e., WFGI-WFG9 and MOKP with 5, 10 and 15 objectives. The results suggest the effectiveness of this approach.

A Dynamic Multi-Objective Evolutionary Algorithm (DMOEA) usually detects a change in an environment and responds to its dynamics, which can lead to new optimal solutions over time. However, in some real problems, correct change detection cannot be guaranteed. The existing methods can miss changes when there is noise in the landscape, or they can yield false positives, demanding an algorithm to respond to a nonexistent new scenario. To handle DMOPs without such detection, a new DMOEA was proposed in which diversity is inserted into the population by a Gaussian Mixture Model-based Local Search (GMM-LS) strategy depending on a new condition based on the HyperVolume metric, triggered independently of occurrences of changes. The parameters of the GMM are determined using Variational Inference. The experiments were performed on FDA1-5 and dMOPl -2, and in a real-world problem. The experimental results suggest the efficacy of the method.

When solving a multi-objective optimization problem using Evolutionary Algorithms, the diversity loss can occur as the evolution process is made. This is particularly significant in Pareto-based strategies where a diversity mechanism is required to maintain a set of solutions well distributed in the Pareto Front extension. Therefore, algorithms are required with the ability to keep a good balance between exploration and exploitation. To address this challenge, a new algorithm is proposed considering past generations to establish trends in population movement, and in this way, to find better Pareto solutions. The proposal, Gaussian Process Regression-based Evolutionary Algorithm (GPR-EA), employs Differential Evolution operators and polynomial mutation. A Gaussian Process model is used to form predictions about the new population in particular generations. The experiments were performed on 15 well-known test functions: UF1-I0 and ZDTI-4, 6. The GPR-EA comparisons with nine algorithms regarding two metrics are presented, evidencing that the proposal outperforms the other algorithms in most problems.

When working with decomposition-based algorithms, an appropriate set of weights might improve quality of the final solution. A set of uniformly distributed weights usually leads to well-distributed solutions on a Pareto front. However, there are two main difficulties with this approach. Firstly, it may fail depending on the problem geometry. Secondly, the population size becomes not flexible as the number of objectives increases. In this paper, we propose the MOEA/D with Uniformly Randomly Adaptive Weights (MOEA/D-URAW) which uses the Uniformly Randomly method as an approach to subproblems generation, allowing a flexible population size even when working with many objective problems. During the evolutionary process, MOEA/D-URAW adds and removes subproblems as a function of the sparsity level of the population. Moreover, instead of requiring assumptions about the Pareto front shape, our method adapts its weights to the shape of the problem during the evolutionary process. Experimental results using WFG41-48 problem classes, with different Pareto front shapes, shows that the present method presents better or equal results in 77.5% of the problems evaluated from 2 to 6 objectives when compared with state-of-the-art methods in the literature.

Evolutionary algorithms have been extensively explored and applied in optimization problems. They allow work with multiple solutions simultaneously, with multimodal functions and dynamic problems, and do not require additional information. Several algorithms have been developed over the years for this task. Yet special attention is needed in the area of increasing the convergence speed of evolutionary algorithms. This study is aimed at developing a framework capable of addressing this new line of research in the field of evolutionary computation. We used the Gaussian Mixture Model to do a local search, and generated a new population through the use of Variational Inference. To implement the proposed framework (GMM-Local Search), NSDE both static and dynamic with multiple objectives were used as basic algorithms. Experiments were performed with different test functions for static and dynamic multi-objective optimization problems. The comparison of the algorithms using the proposed framework with the basic algorithms are presented here, thus evidencing that an improvement in the convergence can be achieved.

Online learning algorithms are intrinsically designed to deal with large amounts of data because of the one-instance-at-a-time approach to the learning process, circumventing memory issues and enabling real time learning. However, most online algorithms require previous knowledge of the problem to predetermine the number of categories to be learned, or some other kind of meta-information that is not likely to be available to a generic system. In this work, an online, incremental algorithm, oiSGNG, is proposed, whose main features are: zero nodes initialization and the original batch SGNG node insertion mechanism [10]. The results improved on the state of the art in 5 out of 12 multiclass datasets.

The probabilistic behavior study of Evolutionary Algorithms (EA) in every generation is relevant to perform exploratory analysis, in order to summarize, monitor, and to formulate a hypothesis about observed data. For the purpose of understanding better how the population evolves along the generations, we made a descriptive analysis of Differential Evolution (DE) evolving population. The objective was to find a probabilistic model to fit the population over the generation. This probabilistic model can be a known probability distribution or a latent variable model, i.e., Gaussian mixture model. In this work, we conducted different adhesion tests for known continuous distributions defined for real variables, namely, Students t, Laplace, and Normal distributions. Among them, the latter showed the highest number of occurrences, hence we conducted a further study over the probability populations distribution based on this result. We used Henze and Zirkler hypothesis tests to verify multivariate normal distribution and a version of the multivariate Kolmogorov test aiming to assess the adjustment to other known multivariate continuous distributions. The probabilistic behavior analysis of the population generated by the DE was made over the single-objective box-constrained continuous optimization problems, the CEC13 benchmarks.

Images, text, web documents, videos, real-world data are very often high-dimensional. Many researchers or users may need to construct accurate predictive models for a variety of applications, especially those that involve clustering. Handling high dimensional data is a reality in processing task involving areas such as high-throughput genotyping platforms and human genetic clustering in bioinformatics, medical imaging and IMRT segmentation in medicine, market research, social network analysis, and anomaly detection. However, the performance of clustering algorithms usually decreases significantly when the sample dimension grows. Moreover, the big data can be acquired taking into consideration different views of them, characterizing the so-called multi-view clustering. In this paper, we use a subspace clustering approach, a time-varying self-organizing map, to deal with multi-view clustering. The method showed itself promising since it can handle real-world data characterized by high sparsity, high dimensionality nature, and different representations. A number of experiments with the proposed solution showed better performance than a number of other state-of-the-art models built specifically to deal with multi-view data.

We introduce a stochastic optimization algorithm, a variation of Opposition-based Multi-objective Differential Evolution with replacement of current population by random individuals in specifics periods, to tune parameters of a Model-Free Adaptive Control System. It is often hard to set such parameters and they are selected according to qualitative analysis of the system response. Evolutionary Algorithms with a single objective function have been used to set the parameters of controllers. Hence, a multi-objective approach was used to solve this problem in order to optimize more than one performance index. In this study, we propose two objective functions in order to approximate the functions of the desired output and the system response. The simulation results suggest the effectiveness of the method to determine the parameters of the controller.

This paper presents an algorithm, self-organizing map-state trajectory generator (SOM-STG), to plan and control legged robot locomotion. The SOM-STG is based on an SOM with a time-varying structure characterized by constructing autonomously close-state trajectories from an arbitrary number of robot postures. Each trajectory represents a cyclical movement of the limbs of an animal. The SOM-STG was designed to possess important features of a central pattern generator, such as rhythmic pattern generation, synchronization between limbs, and swapping between gaits following a single command. The acquisition of data for SOM-STG is based on learning by demonstration in which the data are obtained from different demonstrator agents. The SOM-STG can construct one or more gaits for a simulated robot with six legs, can control the robot with any of the gaits learned, and can smoothly swap gaits. In addition, SOM-STG can learn to construct a state trajectory form observing an animal in locomotion. In this paper, a dog is the demonstrator agent.

Subspace clustering is the task of identifying clusters in subspaces of the input dimensions of a given dataset. Noisy data in certain attributes cause difficulties for traditional clustering algorithms, because the high discrepancies within them can make objects appear too different to be grouped in the same cluster. This requires methods specially designed for subspace clustering. This paper presents our second approach to subspace and projected clustering based on self-organizing maps (SOMs), which is a local adaptive receptive field dimension selective SOM. By introducing a time-variant topology, our method is an improvement in terms of clustering quality, computational cost, and parameterization. This enables the method to identify the correct number of clusters and their respective relevant dimensions, and thus it presents nearly perfect results in synthetic datasets and surpasses our previous method in most of the real-world datasets considered.

In Genome-Wide Association Studies (GWAS) huge amounts of genetic information are analyzed in order to discover how the observed variations, more specifically, the Single Nucleotide Polymorphisms (SNPs), are related with a certain trait of interest, such as the susceptibility for a disease. However, the high dimensionality observed in the datasets imposes significant challenges for methods that try to identify the relevant SNPs and their interactions. In particular, we emphasize the challenges imposed by the great amount of irrelevant dimensions shadowing information which is object of study. In this work, we present a prototype-based classification method, derived from Learning Vector Quantization (LVQ), in which the relevance of each input dimension is learned independently for each prototype. We validate our method in simulated datasets of GWAS with a significant number of dimensions (20, 50, or 100) in which few of them (from 2 to 5) are relevant. Such dimensions have to be identified. The proposed method presented promising results, showing graceful degradation when the number of irrelevant dimensions increases, in comparison with Multifactor Dimensionality Reduction (MDR), Generalized Relevance Learning Vector Quantization (GRLVQ) and Supervised Relevance Neural Gas (SRNG).

High dimensional datasets usually present several dimensions which are irrelevant for certain clusters while they are relevant to other clusters. These irrelevant dimensions bring difficulties to the traditional clustering algorithms, because the high discrepancies within them can make objects appear too different to be grouped in the same cluster. Subspace clustering algorithms have been proposed to address this issue. However, the problem remains an open challenge for datasets with noise and outliers. This article presents an approach for subspace and projected clustering based on Self-Organizing Maps (SOM), that is called Dimensional Selective Self-Organizing Map. DSSOM keeps the properties of SOM and it is able to find clusters and identify their relevant dimensions, simultaneously, during the self-organizing process. The results presented by DSSOM were promising when compared with state of art subspace clustering algorithms.

The insertion of atypical solutions (immigrants) in Evolutionary Algorithms populations is a well studied and successful strategy to cope with the difficulties of tracking optima in dynamic environments in single-objective optimization. This paper studies a probabilistic model, suggesting that centroid based diversity measures can mislead the search towards optima, and presents an extended taxonomy of immigration schemes, from which three immigrants strategies are generalized and integrated into NSGA2 for Dynamic Multiobjective Optimization (DMO). The correlation between two diversity indicators and hypervolume is analyzed in order to assess the influence of the diversity generated by the immigration schemes in the evolution of non-dominated solutions sets on distinct continuous DMO problems under different levels of severity and periodicity of change. Furthermore, the proposed immigration schemes are ranked in terms of the observed offline hypervolume indicator.

This paper reports a study of the influence of diversity in the convergence dynamics of Multiobjective Evolutionary Algorithms (MOEAs) towards the Pareto Front (PF). By varying mutation and crossover parameters, several scenarios of exploration and exploitation are considered, in which each of them is analysed in order to assess the role of diversity levels on the evolution of high quality sets of non-dominated solutions, in terms of the Hypervolume indicator. For this task, the application of the NSGA2 algorithm for approximating the PF in five ZDT benchmark problems is considered. The results not only indicate that there are significant statistical correlations between several diversity metrics and the observed maximum Hypervolume levels on the evolved populations, but also suggest that there are predictable temporal patterns of correlation when the evolutionary process is portrayed generation wise.

In dynamic environments, the absence of diversity may degrade the performance of evolutionary algorithms (EAs). In a previous article, we introduced an method, diversity-reference adaptive control (DRAC), to control population diversity based on reference diversity. DRAC aims to track an appropriate diversity level through a control-based strategy. In such a strategy, the evolutionary process is seen as a control problem, in which the process output is the population diversity and the process input is one or more EA adjustable parameters. In that first version of DRAC, the evolutionary process is treated as a black box, thus, the updating of the control variables is made as a function of the error between the population diversity and the reference-model diversity. The DRAC approach does not consider sensitivity analysis. In the current version, a population dynamics model is used to describe the evolutionary process and to allow the control variables updating.

The Multiobjective Evolutionary Algorithms (MOEAs) are often applied to solve difficult optimization problems, but the dynamic case is even more special. During the optimization, if the environment is changed, a dynamic algorithm must temporarily increase the exploration and decrease the exploitation to generate genetic diversity and then be capable of handling the new behavior of the environment. A technique to increase the diversity may impose an extra delay to such an algorithm that needs to be fast because the new changes may arrive at any time. This paper proposes a model that adds a mutation operator based on gradient, which has the purpose of generating guided diversity to respond to changes in the environment, hence it can accelerate the convergence of the algorithm as a whole. The memetic mutation operator was inserted in the SPEA2 to respond more efficiently to the modifications. Simulations of the proposed model (called Gradient Guided SPEA2, GSPEA2) were carried out for the benchmarks FDA1, FDA3, and DIMP1. Considering the metrics VDweightedand MSweighted, performance of SPEA2 with GSPEA2 was compared with other four dynamic MOEAs. Results suggest that this is a promising approach.

This paper introduces modifications to a surface reconstruction method previously proposed, the GSRM-iDT, to reduce its processing time allowing its use in real world applications. GSRM-iDT is an incremental self-organizing map that produces 2-manifold meshes. The real world application in this paper concerns the reconstruction of human anatomical structures. Experimental results show that the method proposed can produce meshes that approximate very well the shape of the target anatomical structures with a smaller number of vertices then the original object. We compared our results with a well established geometric method (Power Crust) and, according to the metrics considered, the proposed method achieved better results.

We propose a modular neural network model to simulate the occurrence of false memories. The model was built considering some brain structures and functions involved with the phenomenon as well as concepts from the fuzzy-trace theory, such as the differentiation of the gist and verbatim information and the context formation during the memorization process, including it during the storage and recognition processes. A recurrent version of the ART2 is the core of this model. We carried out simulations which suggest that the model can replicate some characteristics of false memory phenomenon observed in experiments with humans.

We propose a surface reconstruction method based on self-organizing maps that grows incrementally. The method, a variant of a previously proposed method called Growing Self Reconstruction Maps (GSRM), produces a triangulation that satisfies the local criterion of a Delaunay triangulation of immersed surfaces in R3. Experimental results show that the proposed method produces meshes as good as its predecessor (GSRM), while producing meshes in which their edges are locally Delaunay.

The growing self-organizing surface map (GSOSM) is a novel map model that learns a folded surface immersed in a 3D space. Starting from a dense point cloud, the surface is reconstructed through an incremental mesh composed of approximately equilateral triangles. Unlike other models such as neural meshes (NM), the GSOSM builds a surface topology while accepting any sequence of sample presentation. The GSOSM model introduces a novel connection learning rule called competitive connection Hebbian learning (CCHL), which produces a complete triangulation. GSOSM reconstructions are accurate and often free of false or overlapping faces. This letter presents and discusses the GSOSM model. It also presents and analyzes a set of results and compares GSOSM with some other models.

In this paper, we propose a new method for surface reconstruction based on growing self-organizing maps (SOMs), called growing self-reconstruction maps (GSRMs). GSRM is an extension of growing neural gas (GNG) that includes the concept of triangular faces in the learning algorithm and additional conditions in order to include and remove connections, so that it can produce a triangular two-manifold mesh representation of a target object given an unstructured point cloud of its surface. The main modifications concern competitive Hebbian learning (CHL), the vertex insertion operation, and the edge removal mechanism. The method proposed is able to learn the geometry and topology of the surface represented in the point cloud and to generate meshes with different resolutions. Experimental results show that the proposed method can produce models that approximate the shape of an object, including its concave regions, boundaries, and holes, if any.

In this article, we propose a new approach for fast recognition of objects from two-dimensional binary images using descriptors of curvature, the moment and an artificial neural network. This model associates a coefficient of certainty for each classification. Two image descriptors where used, the Hu moments and curvature scale space, to provide a reduced representation invariant to image transformations, and a neural network applying a Gibbs distribution of probability is used to calculate the coefficient of certainty to link an image to one class. A benchmark data set is used to demonstrate the usefulness of the proposed methodology. The robustness of the proposed approach is also evaluated under rotation, scale transformations. The evaluation of the performance is based on the accuracy in the framework of a Monte Carlo experiment.

This paper proposes two new data clustering approaches using the particle swarm optimization algorithm (PSO). It is shown how the PSO can be used to find centroids of a user specified number of clusters. The proposed approaches are an attempt to improve the Merwe and Engelbrecht method using different fitness functions and considering the situation where data is uniformly distributed. The data clustering PSO algorithm, using the original and proposed fitness functions is evaluated on well known data sets. Notable improvements on the results were achieved by the modifications, this shows the potential of the PSO, not only on data clustering but also on the several areas it can be applied.

Maintaining the population diversity in genetic algorithms (GAs), or minimize its loss, may benefit the evolutionary process in several ways. The premature convergence may lead the GA to a non-optimal result, that is, converging to a local optimum. Specially in dynamic problems, the diversity preservation is a crucial issue. In this work, a study of different diversity models based on several works has been made. From these models a diversity reference-model has been created in order to enhance diversity-reference adaptive control (DRAC) [20] performance. This new version of DRAC method was evaluated in case studies using a dynamic test functions presented in [26]. The validation of the proposed adaptive parameter control method was performed comparing its performance with SGA and other diversity-based algorithm.

The performance of evolutionary algorithms (EAs) may be enhanced whether the choice of some parameters, as mutation rate and crossover method, is made appropriately. Several methods to adjust those parameters have been developed in order to enhance EAs performance. For this reason, it is important to understand EA dynamics. This paper presents a new population dynamics model to describe and predict the diversity at one generation. The formulation is based on the selection probability density function of each individual. The proposed population dynamics is modeled for an infinite population with generational evolution method. The model was tested in several case studies of different population sizes. The results suggest that the prediction error decreases with the population size increasement.

The collective foraging behavior of ants is an example of self-organization and

This paper presents a new Self-organizing Map suitable for recovering a 2D surface starting from points sampled on the object surface. Growing self-organizing surface map (GSOSM), is a new algorithm of the growing SOM family that reproduce the surface as an incremental mesh composed of triangles which are approximately equilateral. GSOSM introduces a new connection learning rule, called competitive connection Hebbian learning (CCHL), that produces a complete triangulation where CHL fails. Differently from other models such as neural meshes (NM), GSOSM recovers a surface topology from homogeneous samples distribution according to any presentation sequence. GSOSM map is a mesh that represents the object surface with a detail level established by a parameter, allowing different versions of a same object surface. Moreover, GSOSM reconstructions are very often meshes free of false or overlapping faces, and then GSOSM is a potential tool for virtual reconstruction of real objects.

Preservation of diversity in the evolutionary process is crucial to solve problems considering dynamic environments. This work proposes an adaptive evolutionary algorithm to control the population diversity based on a diversity function. The evolutionary process searches for the optimum while the diversity is controlled to track the diversity function. To control the population diversity, the proposed method creates a selection mechanism to adjust the fitnesses of a part of the population based on a fitness penalty. The proposed adaptive method uses the model-reference adaptive system as the control strategy to adjust the fitness penalty parameter. The proposed method is called diversity-reference adaptive control (DRAC). The performance of DRAC method was evaluated for multimodal and dynamic test functions. The results show that DRAC method often reached the optimum area, following environment changes, faster than SGA.

This work introduces a new method for surface reconstruction based on Growing Self-organizing Maps, which learn 3D coordinates of each vertex in a mesh as well as they learn the topology of the input data set. Each map grows incrementally producing meshes of different resolutions, according to the application needs. Another highlight of the presented algorithm refers to the reconstruction time, which is independent from the size of the input data. Experimental results show that the proposed method can produce models that approximate the shape of an object, including its concave regions and holes, if any.

An incremental self-organizing map, called State Trajectory Generator (STRAGEN) is employed to plan state trajectories of a robot. STRAGEN can deal with different criteria to construct topological maps of the problem space, choosing neighbors that match these criteria and optimize different measures of the learned map. STRAGEN can also learn heterogeneous information, such as angles, torques and positions of a manipulator, preserving their characteristics. This algorithm was tested by generating trajectories for a robotic hand called Kanguera. Kanguera presents a new concept of anthropomorphic robot hand. The hand offers a suitable environment for experimental purposes due to its novel and more accurate transmission system. The implementation of adduction and abduction capacity for both the fingers and the thumb allows the execution of more complex movements. Simulations and experiments related to Kanguera hardware are also presented.

This paper presents a self-organized artificial neural network model, called State Trajectory Generator (STRAGEN) capable of generating state trajectories. The model is incremental, it can grow and diminish dynamically during the training phase and adapt itself to the represented space. STRAGEN can consider different criteria to choose neighbors and to adapt to different domains or different characteristics of a same domain. This capacity enables STRAGEN with a representation strategy that can deal with heterogeneous information. Moreover, different criteria also allow STRAGEN to generate trajectories that optimize different measures of the problem space. The algorithm was tested to generate trajectories in a robotic manipulators domain, with two and three dimensions.

Classical approaches of multicast routing consider a tree path whose computational cost entails high use of resources such time and memory in the optimization process. This paper presents a genetic algorithm model applied to the multicast routing problem, in which no tree is built. The solution aims to maximize common paths in source-destinations routes and to minimize the route sizes. New options of fitness functions, variation and selection operators were proposed to increase the ability to generate feasible routes. The simulations were performed in two networks: the 33-node European GEANT WAN network to assess the capacity to find viable solutions and a 100-node network to test the capacity to handle larger networks. The results suggest promising performance for this approach.

This paper presents a multiobjective genetic algorithm to solve the multicast routing problem without using multicast trees. The mechanism to find routes aims to fulfill two conflicting objectives: maximization of the common links in source-destination routes and minimization of the route sizes. The proposed GA can be characterized by representation of network links in a permutation problem, local viability restrictions to generate the initial population with a significant number of feasible routes, variation operators with viability constraints, selection operators to select the most promising and preserve diversity, and fitness function to deal with the conflicting objectives. The model was tested in three networks: the 33-nodes European GEANT WAN network backbone and two networks (66-node and 100-node) randomly generated using the Waxman model at BRITE network topology generator. The multicast results suggest promising performance compared with the unicast shortest path routing.

The family of chaotic bidirectional associative memory (C-BAM family) can reach all stored patterns during the chaotic behavior. Therefore, in this case, C-BAM family can not converge towards a specific pattern, consequently, a desired output is not available. We introduce a control strategy to make heteroassociative chaotic networks converge towards a non-accessible memory and towards a last state of a trajectory. Computer simulations showed that the chaos of C-BAM family could be controlled through the pinning control method, in which any stored output can be recalled from its associated stimulus.

Nonlinear artificial neural network models are very attractive for modeling and forecasting time series. The use of such models in these types of applications is motivated by experimental results that show a high capacity of approximation for functions with high accuracy. However, many researchers have used feedforward and/or backpropagation models for time series predictions. In this paper, a model is applied for neural networks with the dynamic architecture proposed by Ghiassi and Saidane (2005), known as the DAN2 model. The results of DAN2 are compared with auto-regressive integrated mobile average (ARIMA) models. As the main result of the paper, we propose a hybrid model with dynamic architecture (HAD) based on combinations of individual forecasts from the DAN2 and ARIMA models with the aim of obtaining more precise forecasts for poorly behaved time series. The results suggest that for this kind of series, the HAD hybrid model outperforms the individual DAN2 and ARIMA models.

A chaotic bidirectional memory model (C-BAM) is constructed through the inclusion of chaotic neurons in the original BAM. Empiric experiments showed the occurrence of a chaotic dynamic capable to generate large diversity of recalled patterns involving complex excursions over all stored memories. This suggested that the retrieval sequence can model the ability of a novice or the ability of an expert to execute a task. Moreover, the paper illustrates a case in which a novice recall can be transformed into an expert recall through parametric variation.

This work describes the use of direct supervised multi layer perceptron network (MLP) with one hidden layer. Its weights are adjusted by the backpropagation algorithm. In an artificial neural network (ANN), the knowledge of the domain specialists is represented by the topology of the ANN and by the values of the weights used. Thus, it is considerably difficult to explain to a specialist of a domain, how an ANN achieved its outputs. In order to solve this problem, we utilize a rules extraction mechanism, from the trained network, of the kind IF/THEN to explain the results obtained by the network. It is worth noting that such rules are more acceptable by specialists, due to their resemblance to the human reasoning. In order to accomplish this task, a breast cancer database and another with minimum indexes from BOVESPA were adopted to assess the capacity for classification and prediction of the implemented model.

In this paper, we introduce a general modeling technique, called vector-quantized temporal associative memory (VQTAM), which uses Kohonen's self-organizing map (SOM) as an alternative to multilayer perceptron (MLP) and radial basis function (RBF) neural models for dynamical system identification and control. We demonstrate that the estimation errors decrease as the SOM training proceeds, allowing the VQTAM scheme to be understood as a self-supervised gradient-based error reduction method. The performance of the proposed approach is evaluated on a variety of complex tasks, namely: i) time series prediction; ii) identification of SISO/MIMO systems; and iii) nonlinear predictive control. For all tasks, the simulation results produced by the SOM are as accurate as those produced by the MLP network, and better than those produced by the RBF network. The SOM has also shown to be less sensitive to weight initialization than MLP networks. We conclude the paper by discussing the main properties of the VQTAM and their relationships to other well established methods for dynamical system identification. We also suggest directions for further work.

In this paper we proposed an unsupervised neural architecture, called Temporal Parametrized Self Organizing Map (TEPSOM), capable of learning and reproducing complex robot trajectories and interpolating new states between the learned ones. The TEPSOM combines the Self-Organizing NARX (SONARX) network, responsible for coding the temporal associations of the robotic trajectory, with the Parametrized Self-Organizing (PSOM) network, responsible for an efficient interpolation mechanism acting on the SONARX neurons. The TEPSOM network is used to model the inverse kinematics of the PUMA 560 robot during the execution of trajectories with repeated states. Simulation results show that the TEPSOM is more accurate than the SONARX in the reproduction of the learned trajectories.

Spatiotemporal connectionist networks (STCNs) comprise an important class of neural models that can deal with patterns distributed in both time and space. In this article, we widen the application domain of the taxonomy for supervised STCNs recently proposed by Kremer (2001) to the unsupervised case. This is possible through a reinterpretation of the state vector as a vector of latent (hidden) variables, as proposed by Meinicke (2000). The goal of this generalized taxonomy is then to provide a nonlinear generative framework for describing unsupervised spatiotemporal networks, making it easier to compare and contrast their representational and operational characteristics. Computational properties, representational issues, and learning are also discussed, and a number of references to the relevant source publications are provided. It is argued that the proposed approach is simple and more powerful than the previous attempts from a descriptive and predictive viewpoint. We also discuss the relation of this taxonomy with automata theory and state-space modeling and suggest directions for further work.

We investigate dynamic channel assignment (DCA) in mobile communications systems using genetic algorithms (GA). Two new strategies using GA are proposed. In the first strategy, GAL, channels previously assigned are kept locked during the call holding time. In the second strategy, GAS, calls can be switched to different channels during the connection time. We evaluate the performance of the proposed GAs in a 49 hexagonal cell arrangement operating under uniform and nonuniform traffic distributions. Numerical results show that the average call blocking probability of the GAS strategy is lower than that of fixed channel assignment with a borrowing directional channel-locked(BDCL) scheme and of DCA based on Q-learning. The performance of the GAL strategy is better than Q-learning-based DCA for all the investigated cases.

A self-organizing neural net for learning and recall of complex temporal sequences is developed and applied to robot trajectory planning. We consider trajectories with both repeated and shared states. Both cases give rise to ambiguities during reproduction of stored trajectories which are resolved via temporal context information. Feedforward weights encode spatial features of the input trajectories, while the temporal order is learned by lateral weights through delayed Hebbian learning. After training, the net model operates in an anticipative fashion by always recalling the successor of the current input state. Redundancy in sequence representation improves noise and fault robustness. The net uses memory resources efficiently by reusing neurons that have previously stored repeated/shared states. Simulations have been carried out to evaluate the performance of the network in terms of trajectory reproduction, convergence time and memory usage, tolerance to fault and noise, and sensitivity to trajectory sampling rate. The results show that the model is fast, accurate, and robust. Its performance is discussed in comparison with other neural-networks models.

A distributed robot control system is proposed based on a temporal self-organizing neural network, called competitive and temporal Hebbian (CTH) network. The CTH network can learn and recall complex trajectories by means of two sets of synaptic weights, namely, competitive feedforward weights that encode the individual states of the trajectory and Hebbian lateral weights that encode the temporal order of trajectory states. Complex trajectories contain repeated or shared states which are responsible for ambiguities that occur during trajectory reproduction. Temporal context information are used to resolve such uncertainties. Furthermore, the CTH network saves memory space by maintaining only a single copy of each repeated/shared state of a trajectory and a redundancy mechanism improves the robustness of the network against noise and faults. The distributed control scheme is evaluated in point-to-point trajectory control tasks using a PUMA 560 robot. The performance of the control system is discussed and compared with other unsupervised and supervised neural network approaches. We also discuss the issues of stability and convergence of feedforward and lateral learning schemes.

A distributed robot control system is proposed based on a temporal self-organizing neural network, called a competitive temporal Hebbian (CTH) network. The CTH network can learn and recall complex trajectories using two sets of synaptic weights, namely competitive feedforward weights that encode the individual states of the trajectory and Hebbian lateral weights that encode the temporal order of the trajectory states. Ambiguities that occur during trajectory reproduction are resolved using temporal context information. Also, the CTH network saves memory space by maintaining only a single copy of each repeated/shared state of a complex trajectory. A distributed processing scheme is proposed to evaluate the CTH network in the point-to-point real-time trajectory control of a Puma 560 robot. The performance of the control system is discussed and compared with other neural network approaches.

An unsupervised neural network is proposed to learn and recall complex robot trajectories. Two cases are considered: (i) A single trajectory in which a particular arm configuration (state) may occur more than once, and (ii) trajectories sharing states with each other. Ambiguities occur in both cases during recall of such trajectories. The proposed model consists of two groups of synaptic weights trained by competitive and Hebbian learning laws. They are responsible for encoding spatial and temporal features of the input sequences, respectively. Three mechanisms allow the network to deal with repeated or shared states: local and global context units, neurons disabled from learning, and redundancy. The network reproduces the current and the next state of the learned sequences and is able to resolve ambiguities. The model was simulated over various sets of robot trajectories in order to evaluate learning and recall, trajectory sampling effects and robustness.

A self-organizing neural network for learning and recall of complex temporal sequences is proposed. We consider a single open or closed sequence with repeated items, or several sequences with a common state. Both cases give rise to ambiguities during recall of such sequences which is resolved through context input units. Competitive weights encode spatial features of the input sequence, while the temporal order is learned by lateral weights through a time-delayed Hebbian learning rule. Repeated or shared items are stored as a single copy resulting in an efficient memory use. In addition, redundancy in item representation improves the network robustness to noise and faults. The model operates by recalling the next state of the learned sequences and is able to solve potential ambiguities. The model is simulated with binary and analog sequences and its functioning is compared to other neural networks models.

An unsupervised neural network that uses Hebbian and anti-Hebbian learning (HAHL model) was implemented to determine levels of anxiety of rats by clustering these animals based on their behavior in the elevated plus maze. The HAHL model showed capacity to generalize, being trained with only 1.6 of the total of patterns, and was able to identify fine details during the clustering, i.e. sensibility to context and scale. Analysis of the results showed that the proposed model was able to coherently cluster the animals in different exploratory activities, and consequently, in different levels of anxiety.

A self-organizing neural network is proposed to handle multiple temporal sequences with states in common. The proposed network combines context-based competitive learning with time-delayed Hebbian learning to encode spatial features and temporal order of sequence items. A responsibility function to avoid catastrophic forgetting, and a redundancy mechanism to provide noise and fault tolerance increase the reliability of the model. States shared by different sequences are encoded by a single neuron, whereas context information indicates the correct sequence to be recalled in the case of ambiguity. Simulations with trajectories of a PUMA 560 robot are performed to test the network accuracy, robustness to noise and tolerance to faults.

Presents a neural system to determine three contact points between a gripper and an object of arbitrary shape. The neural system is composed of three functional blocks to capture and process the image, establish the contact points and estimate the contact forces. The second block is formed by two neural networks. The first network (competitive Hopfield neural network) determines an approximate polygon for an object outline. A second network, a RBF or MLP model, defines three contact points. The results suggest that the neural system always reaches stable grasping for known and unknown objects Moreover, the training methods used by the RBF model influences significantly the performance and the learning speed of the system.

This paper presents a two-stage neural system to determine the contact points between a three-fingered gripper and an object of arbitrary shape. In the first stage, a CCD camera captures the image of the object and such an image is transformed into a two-dimensional outline through a nearest neighbour algorithm. In the second phase, two neural networks, functioning in cascade, select three contact points in the outline. A competitive Hopfield neural network defines an approximate polygon considering a reduced number of boundary points of the original outline. Then, a supervised neural network, either a multi-layer perceptron or a radial basis function (RBF) network, find the contact points. The experiments suggest that the RBF network trained by the global ridge regression method is suitable for on-line applications and presents the best overall performance in terms of accuracy and robustness to noise. Moreover, this method is able to find correctly the contact points for objects of arbitrary shapes.

This paper describes an unsupervised neural network model for learning and recall of temporal patterns. The model comprises two groups of synaptic weights, named competitive feedforward and Hebbian feedback, which are responsible for encoding the static and temporal features of the sequence respectively. Three additional mechanisms allow the network to deal with complex sequences: context units, a neuron commitment equation, and redundancy in the representation of sequence states. The proposed network encodes a set of robot trajectories which may contain states in common, and retrieves them accurately in the correct order. Further tests evaluate the fault-tolerance and noise sensitivity of the proposed model.

This paper describes a reinforcement learning algorithm that allows a point robot to learn navigation strategies within initially unknown indoor environments with fixed and dynamic obstacles. The knowledge is encoded in two surfaces, called reward and penalty surfaces, that are updated either when a target is found or whenever the robot moves respectively. The proposed policy is suitable for both planning and reactive behaviour. The tests involve different kinds of obstacles: a fixed passage, a barrier, a U-shape obstacle and a simple maze. The results suggest that the model solves the goal-directed exploration problem. Thus, the robot is able to reach a desired goal, starting its movement from any position within the environment, avoiding obstacles, and following a viable trajectory. The robot may get stuck in dynamic obstacles, may depend on randomness to avoid them, and generally does not solve the goal-directed reinforcement learning problem.

This paper proposes a three-layer partially recurrent neural network to perform trajectory planning, solve the inverse kinematics and the inverse dynamics problems in a single processing stage. The feedforward structure of the neural model entails fully connected layers. The feedback links consists in output-input and input-input connections. All the connections are trainable by error backpropagation with variable learning rate and momentum. The network generated trajectories for the PUMA 560 manipulator. The tests comprise generation of four different types of trajectories. Each path is provided in spatial positions, joint angles and joint torques. The results suggest that the model is able to yield trained trajectories given only their initial and final points. Moreover, the results suggest that the model is robust to noise in the trajectories with lower level of complexity.