International Journal of Computational Intelligence and Applications

The article outlines an evolutionary account relating motivational systems to emotions and the cortical systems which elaborate them. It then addresses the issue of whether and how to characterise emotions in such a way that one might say that a robot has emotions even if they are not empathically linked to human emotions.

For face recognition from video streams speed and accuracy are vital aspects. The first decision whether a preprocessed image region represents a human face or not is often made by a feed-forward neural network (NN), e.g. in the Viisage-FaceFINDER^® video surveillance system. We describe the optimisation of such a NN by a hybrid algorithm combining evolutionary multi-objective optimisation (EMO) and gradient-based learning. The evolved solutions perform considerably faster than an expert-designed architecture without loss of accuracy. We compare an EMO and a single objective approach, both with online search strategy adaptation. It turns out that EMO is preferable to the single objective approach in several respects.

Selecting parameters can be a powerful mechanism in constructing new evolving connectionist network. However, if a parameter contains partial information such that only some of the values are relevant and others are not, then a selection of the subset of relevant values is more appropriate. Considering the possible values of a parameter of a processing connectionist network as the outcomes of a variable, this research focuses on selecting interval values of the variable. It also considers the partitioning schemes used in generating the intervals from the outcomes of a variable. The goal of this work is to explore variable value selection and its effect in an evolving connectionist network. Using input variables in a backpropagation network, the proposed method evaluates its effect based on training of a dataset, and eliminates those intervals of the variable values that contribute negatively when processed by the network. When a value falls into an interval that has been selected and ignored, it is analogous to a network without processing the corresponding variable, and vice versa. Two approaches for interval partitioning are considered, based on equal-probability (or maximum entropy) and equal-width partitioning scheme. Comparing the best performing network with selection and the one without selection, the experimental results show that the best network with selection can produce better performance accuracy and smaller network size.

In the paper a special incremental identification procedure for learning a Neural Singleton (NS) Model, as a simplification of the RBFNN model is proposed. At each epoch of this incremental procedure a special integrated learning algorithm for the neuron centers and then a supervised learning for the singletons are performed. As a result, the approximation error of the NS model is gradually decreased with the increment of the neurons at each learning epoch.

The newly proposed integrated unsupervised and Error-Driven learning algorithm is the main part of the incremental identification procedure. It is a modified version of the Neural-Gas learning algorithm, which uses the normalised evaluation error as feedback information. Thus the integrated learning algorithm is able to drive the neurons to the "more interesting" areas in the input space, where bigger evaluation errors exist, instead of just locating them in the areas of higher data density. The final result is an incremental growing type of NS model that gradually improves its approximation accuracy at each learning epoch.

A detailed test example is shown in the paper in order to evaluate the performance and to show the important features of the whole incremental identification procedure, including the integrated learning algorithm. Comparison results with the conventional (one-epoch and non-incremental) model are also given.

Emergence of game strategy in multiagent systems is studied. Symbolic and subsymbolic (neural network) approaches are compared. Symbolic approach is represented by a backtrack algorithm with specified search depth, whereas the subsymbolic approach is represented by feedforward neural networks that are adapted by reinforcement temporal difference TD(λ) technique. As a test game, we use simplified checkers. The problem is studied in the framework of multiagent system, where each agent is endowed with a neural network used for a classification of checkers positions. Three different strategies are used. The first strategy corresponds to a single agent that repeatedly plays games against MinMax version of a backtrack search method. The second strategy corresponds to single agents that are repeatedly playing a megatournament, where each agent plays two different games with all other agents, one game with white pieces and the other game with black pieces. After finishing each game, both agents modify their neural networks by reinforcement learning. The third strategy is an evolutionary modification of the second one. When a megatournament is finished, each agent is evaluated by a fitness, which reflects its success in the given megatournament (more successful agents have greater fitness). It is demonstrated that all these approaches lead to a population of agents very successfully playing checkers against a backtrack algorithm with the search depth 3.

A method for extracting Zadeh–Mamdani fuzzy rules from a minimalist constructive neural network model is described. The network contains no embedded fuzzy logic elements. The rule extraction algorithm needs no modification of the neural network architecture. No modification of the network learning algorithm is required, nor is it necessary to retain any training examples. The algorithm is illustrated on two well known benchmark data sets and compared with a relevant existing rule extraction algorithm.

This work applies Evolutionary Computation to achieve completely self-adapting Evolving Fuzzy Neural Networks (EFuNNs) for operating in both incremental (on-line) and batch (off-line) modes. EFuNNs belong to a class of Evolving Connectionist Systems (ECOS), capable of performing clustering-based, on-line, local area learning and rule extraction. Through Evolutionary Computation, its parameters such as learning rates and membership functions are continuously adjusted to reflect the changes in the dynamics of incoming data. The proposed methods are tested on the Mackey–Glass series and the results demonstrate a substantial improvement in EFuNN's performance.