This paper reports on using qualitative data analysis to understand aspects of the well-being of dwellers in underserved communities, by applying machine learning algorithms to identify specific themes from unstructured interview data. The work involved data translation, transcription, pre-processing as well as developing Word2Vec and FastText algorithms and ultimately a combined analysis engine. The reported experiments are conducted on field data captured from communities in India, hence offering a unique opportunity to examine automated context-based qualitative data analysis. The approach is proven feasible despite the dominant limitations on technology infrastructure and community awareness. The machine learning results identify themes from the interview data within minutes as opposed to hours of manual investigations through conventional qualitative analysis techniques. The outcomes from the analysis engine can be used for creating a grounded theory for further studies, hence facilitating an evidence-based approach to the evaluation of underserved communities.

This paper reports on the commercialization of a business intelligence application deploying computational intelligence techniques. Theoretical foundations are included where appropriate, along with implementation and comparative benchmark results. Discussions on technology transfer mechanisms are included, identifying a generic framework for the commercialization of technology innovations, with a particular case study from Jordan.

The semiconductor manufacturing domain is by no doubt a rich and challenging environment for the application of machine learning. Some of the demanding characteristic of semiconductor data include high dimensionality, mixtures of categorical and numerical data, non-randomly Gaussian data, non-Gaussian and multi-modal distributions, highly non-linear complex relationships, noise and outliers in both x and y dimensions, temporal dependencies, etc. These challenges are becoming particularly crucial as the quantity of available data is growing dramatically. This paper addresses the problem of automatic wafer manufacturing process error detection based on electrical wafer sort (EWS) parametric tests. A wafer-to-wafer analysis is presented that automatically detect possible errors in the manufacturing process that causes systematic damage to the product as it passes through some step in the process. Possible causes are equipment mishandling, operator error, material issues such as contamination, etc. These manufacturing errors are reflected on the physical and electrical properties of the wafers, which are measured by the parametric tests in the EWS process. The core of this research is a novel classifier system based on the benefits arising from the interaction between evolutionary algorithms and artificial neural networks. Experimental results demonstrates that this system is able to detect defective wafers with an accuracy of 82%. Prior to the EWS classification, the proposed system is evaluated with three classification benchmark problems: Iris dataset, Australian credit card problem, and the Pumas diabetes dataset. The obtained results are compared with classifier's outcomes available in the literature.

An automatic defect classification system is introduced for electrical test analysis of semiconductor wafer using evolutionary algorithm techniques to construct radial basis function neural networks (RBF NNs) as a classifier. The parameters of a RBF NN (number of neurons, and their respective centers and radii) are often determined by hand or based on methods highly dependent on initial values. In this work, particle swarm optimization algorithm is implemented to build a RBF NN that solves this specific problem. As a primary input source to the network, the system employs electrical binmaps obtained from the test stage of the manufacturing process. To accomplish this task, a filtering algorithm is also implemented able to discard those wafermaps without pattern. The performance of the reported approach shows an outstanding e-bitmap classification rate. To evaluate the performance of the main algorithm, the system is tested also on the Australian credit card data set and the error rate obtained is comparable with the best algorithms found in the literature.

An automatic defect classification system for electrical test analysis of semiconductor wafer using a combination of a genetic algorithm as a feature selector and a RBF neural network as a classifier is introduced. An e-bitmap is obtained from the test stage and several features including mass, moments and invariant moments are extracted. These are presented to the feature selection method to generate an optimal set of features thus presented to the classifier. The performance of the reported approach show an 87% correct e-bitmap classification rate. The use of features gives this approach the capability to be device independent (i.e. independent of the format of die layout on a tested wafer).

We propose a new evolutionary algorithm for subspace clustering in very large and high-dimensional databases. The design includes task-specific coding and genetic operators, along with a nonrandom initialization procedure. Experimental results show that the algorithm scales almost linearly with the size and dimensionality of the database as well as the dimensionality of the hidden clusters. Our algorithm is able to discover clusters of different densities embedded in both low and high dimensional subspaces of the original space. Finally, the discovered knowledge is presented in the form of nonoverlapping clustering rules where only those features relevant to the clustering are reported. These two properties contributes to the relatively high comprehensibility of the clustering output.

The main objective of this paper is to use genetic algorithms in order to improve the quality of the bathymetry derived from sidescan raw data. The optimisation sequence starts with inverse modelling of the phase data, which uniquely corresponds to the characteristics of the coupled system of the sidescan vehicle and the seafloor terrain. These phase data are then compared with phase data actually collected by the sonar, to produce a correlation coefficient as an objective function. Simulation results are reported for the algorithm showing robust convergence towards the optimum value of the objective function. The results indicate that this new approach can be used to avoid difficulties widely encountered during forward processing of phase data to derive bathymetry.

This paper presents initial investigations into an evolutionary neural network suitable for gait analysis of human motion. The approach here is to develop an intelligent black box that can take the physiological signals (EMG) and interpret them to give accurate information on the position and movement of the knee (gait). Two MLP networks are presented with weight evolving algorithms employing mutation and crossover separately. Simulation results show the evolutionary algorithms exhibiting particularly better ability to generalise solutions, which is an important aspect for the reliability of EMG/gait map generation.

Clustering is a hard combinatorial problem and is defined as the unsupervised classification of patterns. The formation of clusters is based on the principle of maximizing the similarity between objects of the same cluster while simultaneously minimizing the similarity between objects belonging to distinct clusters. This paper presents a tool for database clustering using a rule-based genetic algorithm (RBCGA). RBCGA evolves individuals consisting of a fixed set of clustering rules, where each rule includes d non-binary intervals, one for each feature. The investigations attempt to alleviate certain drawbacks related to the classical minimization of square-error criterion by suggesting a flexible fitness function which takes into consideration, cluster asymmetry, density, coverage and homogeny.

This paper presents the use of neural networks and a genetic algorithm in time-optimal control of a closed-loop 3-dof robotic system. Extended Kohonen networks which contain an additional lattice of output neurons are used in conjunction with PID controllers in position control to minimize command tracking errors. The extended Kohonen networks are trained using reinforcement learning where the overall learning algorithm is derived from a self-organizing feature-mapping algorithm and a delta learning rule. The results indicate that the extended Kohonen network controller is more efficient than other techniques reported in early literature when the robot is operated under normal conditions. Subsequently, a multi-objective genetic algorithm (MOGA) is used to solve an optimization problem related to time-optimal control. This problem involves the selection of actuator torque limits and an end-effector path subject to time-optimality and tracking error constraints. Two chromosome coding schemes are explored in the investigation: Gray and integer-based coding schemes. The results suggest that the integer-based chromosome is more suitable at representing the decision variables. As a result of using both neural networks and a genetic algorithm in this application, an idea of a hybridization between a neural network and a genetic algorithm at the task level for use in a control system is also effectively demonstrated.

This paper presents the use of neural networks and a genetic algorithm within a model-based friction compensation scheme in a closed-loop robotic system. Unlike previous works which concentrate on using a single type of neural network to model friction function, three types of neural network, namely a multilayer perceptron, a radial-basis function network and a modular network, are used in the modelling task. The genetic algorithm is then used to find the optimal combination between the radial-basis function network and the multilayer perceptron and that between the radial-basis function network and the modular network. The simulation results indicate that the genetic algorithm has successfully found the combinations of the neural networks which results in a better compensation performance than using only one type of neural network. As a result of using both neural networks and a genetic algorithm in this application, an idea of a task hybridisation between neural networks and a genetic algorithm for use in a control system is also effectively demonstrated.

This paper introduces a hybrid neural structure using radial-basis functions (RBF) and multilayer perceptron (MLP) networks. The hybrid network is composed of one RBF network and a number of MLPs, and is trained using a combined genetic/unsupervised/supervised learning algorithm. The genetic and unsupervised learning algorithms are used to locate the centres of the RBF part in the hybrid network. In addition, the supervised learning algorithm, based on a back-propagation algorithm, is used to train the connection weights of the MLP part in the hybrid network. Performances of the hybrid network are initially tested using a two-spiral benchmark problem. Several simulation results are reported for applying the algorithm in the classification of myoelectric or electromyographic (EMG) signals where the GA-based network proved most efficient.

The use of intelligent techniques in the manufacturing field has been growing the last decades due to the fact that most manufacturing optimization problems are combinatorial and NP hard. This paper examines recent developments in the field of evolutionary computation for manufacturing optimization. Significant papers in various areas are highlighted, and comparisons of results are given wherever data are available. A wide range of problems is covered, from job shop and flow shop scheduling, to process planning and assembly line balancing.

Truly autonomous vehicles will require both projective planning and reactive components in order to perform robustly. Projective components are needed for long term planning and re-planning where explicit reasoning about future states is required. Reactive components allow the system to always have some action available in real time, and themselves can exhibit robust behaviour, but lack the ability to explicitly reason about future states over a long time period. The paper emphasises creating the projective component but also offer a simple solution for reactive component. A genetic algorithm implements the projective component, which designs automatically a fuzzy logic controller by modifying the position of controller membership functions and the commands given to the robot. For the reactive component, a simple solution was adopted so that if the robot sensors detect new obstacles, the robot will try to move to a previous position.

This paper presents the use of neural networks and genetic algorithms in time-optimal control of a closed-loop robotic system. Radial-basis function networks are used in conjunction with PID controllers in an independent joint position control to reduce tracking errors. The results indicate that using neural network controllers is more effective than using the trajectory pre-shaping scheme, reported in early literature. Subsequently, a genetic algorithm with a weighted-sum approach and a multi-objective genetic algorithm (MOGA) are used to solve a multi-objective optimisation problem related to time-optimal control. The results indicate that the MOGA is the best method in terms of the Pareto front coverage while the genetic algorithm with a weighted-sum approach is more effective in terms of finding the best individual according to the weighted-sum criteria. As a result of using both neural networks and genetic algorithms in this application, an idea of a task hybridisation between neural networks and genetic algorithms for use in a control system is also effectively demonstrated.

Genetic programming has rarely been applied to manufacturing optimisation problems. In this report we investigate the potential use of genetic programming for the solution of the one-machine total tardiness problem. Combinations of dispatching rules are employed as an indirect way of representing permutations within a modified genetic programming framework. Hybridisation of genetic programming with local search techniques is also introduced, in an attempt to improve the quality of solutions. All the algorithms are tested on a large number of benchmark problems with different levels of tardiness and tightness of due dates.

MTGP is a new genetic programming system that uses the multithreading technology of the Java programming language for the parallel evolution of subpopulations of programs. The system runs as a Java applet within a standard web browser on a desktop PC, and utilises a linear program representation for a stack-based virtual machine. The individuals from four subpopulations are manipulated concurrently and these subpopulations exchange their best individuals at regular intervals during a run. MTGP incorporates novel variations on the traditional genetic operators used in genetic programming and in the inclusion of a 'do nothing' gene, in an attempt to produce better evolutionary performance. The basic procedures of the system will be used in the future development of a distributed, Internet-based genetic programming system that will provide large computational power needed to solve complex problems. In this report, the performance of MTGP on two symbolic regression problems is compared to that of four other genetic programming systems. MTGP shows improvement over these systems in terms of the computational effort needed to solve the problems and the accuracy of the solution produced.

The search for the minimum-time path of a robotic manipulator graphically by tessellating the joint space involves heavy computational burden. Genetic algorithms (GAs) have been used to tackle this problem and have reduced the computational search time a great deal. The work presented here provides a practical GAs motion control for possible real-time implementations. A kind of heuristic search technique is found to be necessary to further reduce the search time. The end-velocity is dealt with by placing a penalty on the objective function. A scheme using a fifth order polynomial function to generate smooth initial path population is presented. Much better results with much less cost have been acquired than many of those reported earlier in the literature. Time-optimal motion for a six DOF robot arm is also included, which is made possible by using GAs. It has been found that the quickest motion is not necessarily a straight line in joint space.