Mario Augusto Gongora

Temporal fuzzy association rule mining with 2-tuple linguistic representation

This paper reports on an approach that contributes towards the problem of discovering fuzzy association rules that exhibit a temporal pattern. The novel application of the 2-tuple linguistic representation identifies fuzzy association rules in a temporal context, whilst maintaining the interpretability of linguistic terms. Iterative Rule Learning (IRL) with a Genetic Algorithm (GA) simultaneously induces rules and tunes the membership functions. The discovered rules were compared with those from a traditional method of discovering fuzzy association rules and results demonstrate how the traditional method can loose information because rules occur at the intersection of membership function boundaries. New information can be mined from the proposed approach by improving upon rules discovered with the traditional method and by discovering new rules.

Analysis and test of efficient methods for building recursive deterministic perceptron neural networks

The Recursive Deterministic Perceptron (RDP) feed-forward multilayer neural network is a generalisation of the single layer perceptron topology. This model is capable of solving any two-class classification problem as opposed to the single layer perceptron which can only solve classification problems dealing with linearly separable sets. For all classification problems, the construction of an RDP is done automatically and convergence is always guaranteed. Three methods for constructing RDP neural networks exist: Batch, Incremental, and Modular. The Batch method has been extensively tested and it has been shown to produce results comparable with those obtained with other neural network methods such as Back Propagation, Cascade Correlation, Rulex, and Ruleneg. However, no testing has been done before on the Incremental and Modular methods. Contrary to the Batch method, the complexity of these two methods is not NP-Complete. For the first time, a study on the three methods is presented. This study will allow the highlighting of the main advantages and disadvantages of each of these methods by comparing the results obtained while building RDP neural networks with the three methods in terms of the convergence time, the level of generalisation, and the topology size. The networks were trained and tested using the following standard benchmark classification datasets: IRIS, SOYBEAN, and Wisconsin Breast Cancer. The results obtained show the effectiveness of the Incremental and the Modular methods which are as good as that of the NP-Complete Batch method but with a much lower complexity level. The results obtained with the RDP are comparable to those obtained with the backpropagation and the Cascade Correlation algorithms.

Current trends on knowledge extraction and neural networks

The extraction of knowledge from trained neural networks provides a way for explaining the functioning of a neural network. This is important for artificial networks to gain a wider degree of acceptance. An increasing amount of research has been carried out to develop mechanisms, procedures and techniques for extracting knowledge from trained neural networks. This publication presents some of the current research trends on extracting knowledge from trained neural networks.

Real-time evolution of an embedded controller for an autonomous helicopter

In this paper we evolve the parameters of a proportional, integral, and derivative (PID) controller for an unstable, complex and nonlinear system. The individuals of the applied genetic algorithm (GA) are evaluated on the actual system rather than on a simulation of it, thus avoiding the ldquoreality gaprdquo. This makes implicit a formal model identification for the implementation of a simulator. This also calls for the GA to be approached in an unusual way, where we need to consider new aspects not normally present in the usual situations using an unnaturally consistent simulator for fitness evaluation. Although elitism is used in the GAs, no monotonic increase in fitness is exhibited by the algorithm. Instead, we show that the GApsilas individuals converge towards more robust solutions.

Robustness analysis of evolutionary controller tuning using real systems

A genetic algorithm (GA) presents an excellent method for controller parameter tuning. In our work, we evolved the heading as well as the altitude controller for a small lightweight helicopter. We use the real flying robot to evaluate the GAs individuals rather than an artificially consistent simulator. By doing so we avoid the “reality gap”, taking the controller from the simulator to the real world. In this paper we analyze the evolutionary aspects of this technique and discuss the issues that need to be considered for it to perform well and result in robust controllers.

Evolving temporal fuzzy itemsets from quantitative data with a multi-objective evolutionary algorithm

We present a novel method for mining itemsets that are both quantitative and temporal, for association rule mining, using multi-objective evolutionary search and optimisation. This method successfully identifies temporal itemsets that occur more frequently in areas of a dataset with specific quantitative values represented with fuzzy sets. Current approaches preprocess data which can often lead to a loss of information. The novelty of this research lies in exploring the composition of quantitative and temporal fuzzy itemsets and the approach of using a multi-objective evolutionary algorithm. This preliminary work presents the problem, a novel approach and promising results that will lead to future work. Results show the ability of NSGA-II to evolve target itemsets that have been augmented into synthetic datasets. Itemsets with different levels of support have been augmented to demonstrate this approach with varying difficulties.

A generalised type-2 fuzzy logic system embedded board and integrated development environment

This paper describes the design and construction of the first hardware running a generalised type-2 fuzzy logic system. A rationale for the design is given and hardware representations are discussed. An integrated development environment, also developed as part of this project, for type-2 fuzzy system is described along with the software which links this IDE to the novel hardware. The performance of the novel type-2 development board is tested under two scenarios.

Application of Artificial Neural Network and Support Vector Regression in cognitive radio networks for RF power prediction using compact differential evolution algorithm

Cognitive radio (CR) technology has emerged as a promising solution to many wireless communication problems including spectrum scarcity and underutilization. To enhance the selection of channel with less noise among the white spaces (idle channels), the a priory knowledge of Radio Frequency (RF) power is very important. Computational Intelligence (CI) techniques cans be applied to these scenarios to predict the required RF power in the available channels to achieve optimum Quality of Service (QoS). In this paper, we developed a time domain based optimized Artificial Neural Network (ANN) and Support Vector Regression (SVR) models for the prediction of real world RF power within the GSM 900, Very High Frequency (VHF) and Ultra High Frequency (UHF) FM and TV bands. Sensitivity analysis was used to reduce the input vector of the prediction models. The inputs of the ANN and SVR consist of only time domain data and past RF power without using any RF power related parameters, thus forming a nonlinear time series prediction model. The application of the models produced was found to increase the robustness of CR applications, specifically where the CR had no prior knowledge of the RF power related parameters such as signal to noise ratio, bandwidth and bit error rate. Since CR are embedded communication devices with memory constrain limitation, the models used, implemented a novel and innovative initial weight optimization of the ANNs through the use of compact differential evolutionary (cDE) algorithm variants which are memory efficient. This was found to enhance the accuracy and generalization of the ANN model.

Evolving temporal fuzzy association rules from quantitative data with a multi-objective evolutionary algorithm

A novel method for mining association rules that are both quantitative and temporal using a multi-objective evolutionary algorithm is presented. This method successfully identifies numerous temporal association rules that occur more frequently in areas of a dataset with specific quantitative values represented with fuzzy sets. The novelty of this research lies in exploring the composition of quantitative and temporal fuzzy association rules and the approach of using a hybridisation of a multi-objective evolutionary algorithm with fuzzy sets. Results show the ability of a multi-objective evolutionary algorithm (NSGA-II) to evolve multiple target itemsets that have been augmented into synthetic datasets.

Evolutionary algorithms and fuzzy sets for discovering temporal rules

Abstract A novel method is presented for mining fuzzy association rules that have a temporal pattern. Our proposed method contributes towards discovering temporal patterns that could otherwise be lost from defining the membership functions before the mining process. The novelty of this research lies in exploring the composition of fuzzy and temporal association rules, and using a multi-objective evolutionary algorithm combined with iterative rule learning to mine many rules. Temporal patterns are augmented into a dataset to analyse the method’s ability in a controlled experiment. It is shown that the method is capable of discovering temporal patterns, and the effect of Boolean itemset support on the efficacy of discovering temporal fuzzy association rules is presented.