Lucas Baggio Figueira

Evaluating the effects of distance metrics on a NGE-based system

The nested generalized exemplar (NGE) model (implemented by EACH algorithm) is an incremental form of inductive learning from examples that generalizes a given training set into hypotheses represented as a set of hyper-rectangles in an n-dimensional Euclidean space. NGE depends heavily on the distance metric used in both processes, learning and classification. This work investigates the impact on the predictive accuracy of the learnt concepts by NGE as a consequence of using three new heterogeneous distance functions namely HVDM, IVDM and WVDM, instead of the Euclidean distance metric originally proposed. The paper presents and analyses the results of experiments in various domains using the Euclidean and the three heterogeneous distance functions.

Transferring symbolic knowledge into a RuleNet neural network

This paper investigates the use of symbolic knowledge to initialize a RuleNet neural network learning process. The basic idea is to provide the symbolic knowledge induced by either ID3 or NGE to the RuleNet learning process, as the initial knowledge available. The paper describes experiments conducted using seven knowledge domains and focuses the discussion on the accuracy of the induced concepts. The results show that the collaboration is feasible but not necessarily improves the results obtained by RuleNet on its own

Transferring neural network based knowledge into an exemplar-based learner

This paper investigates knowledge transfer from a neural network based system into an exemplar-based learning system. In order to examine the possibilities of such transfer, it proposes and evaluates a system that implements a collaborative scheme, where a particular type of neural network induced by the neural system RuleNet is used by an exemplar-based system (NGE) to carry on a learning task. The proposed collaboration between the two learning models implemented as the hybrid system RuleNet→NGE is feasible due to the similarity of the concept description languages employed by both. The paper also describes a few experiments conducted; results show that the RuleNet-NGE collaboration is plausible and, in some domains, it improves the performance of NGE on its own.

USING CONSTRUCTIVE NEURAL NETWORKS FOR DETECTING CENTRAL VESTIBULAR SYSTEM LESION

Traditional neural network algorithms such as Backpropagation, require the definition of the network architecture prior to training. Generally these methods work well only when the network architecture is chosen appropriately. It is well known that there is no general answer to the problem of defining a neural network architecture for a given application. This fact has been one of the main motivations for the proposal of constructive neural network algorithms, which try to solve the problem by building the architecture of the neural network during its training. In this paper, we explore the use of three constructive supervised neural network algorithms in learning to identify the possibility of an existing problem in the central vestibular system of a patient, using data from optokinetic tests. The constructive neural network algorithms used are two perceptron-based algorithms, namely, Tower and Pyramid and the DistAl algorithm. The goal of the experiments, aside from creating a good classifier for detecting the presence of a central vestibular system lesion, is also to compare and contrast the performance of the recently proposed DistAl against two well-known constructive algorithms in a medical domain that has not been greatly explored with these types of experiments. In addition, results obtained with Backpropagation are presented for comparison.

Initializing an exemplar based learning process from a RuleNet network

This paper proposes and evaluates a hybrid system based on two machine learning approaches, a neural network and an instance based method. It describes how the knowledge induced by a RuleNet neural network can be used as the initial knowledge for an NGE-like system to start learning. An NGE-based system can be considered an instance based learning method which allows generalization. The proposed collaboration between the two learning methods implemented by the hybrid system is feasible due to the similarity of the concept description languages employed by both. The paper also describes a few experiments conducted; results show that the RuleNet-NGE collaboration is plausible and, in some domains, it improves the performance of NGE on its own.

Choosing the initial set of exemplars when learning with an NGE-based system

In the original proposal of the NGE (nested generalized exemplar) system, the induction of a concept is based on an initial set of training examples (named seeds) that are randomly chosen. The number of examples in this set is arbitrary, generally determined by the user of the system. It can be seen empirically, that the final results are influenced by the initial choice of the seeds. We propose and investigate other alternative methods for choosing seeds and empirically evaluate their impact on the learning results in seven knowledge domains, as far as accuracy and number of expressions describing the concepts are concerned. In spite of the additional time investment associated with using a clustering method and, assuming that accuracy of the induced concept is of major importance, experiments have shown that choosing the initial set of seeds as the center of clusters can be the best option.

Pattern recognition using a recurrent neural network inspired on the olfactory bulb

The olfactory system is a remarkable system capable of discriminating very similar odorant mixtures. This is in part achieved via spatio-temporal activity patterns generated in mitral cells, the principal cells of the olfactory bulb, during odor presentation. In this work, we present a spiking neural network model of the olfactory bulb and evaluate its performance as a pattern recognition system with datasets taken from both artificial and real pattern databases. Our results show that the dynamic activity patterns produced in the mitral cells of the olfactory bulb model by pattern attributes presented to it have a pattern separation capability. This capability can be explored in the construction of high-performance pattern recognition systems.

Learning fuzzy hyper-rectangles with instance and neural based methods

The NGE model is an instance-based inductive learning method that generalizes a given training set into hypotheses represented as a set of hyper-rectangles in a n-dimensional Euclidean space. The RuleNet model does exactly the same thing, but using a neural network algorithm. This paper focuses on a fuzzy version of both algorithms aiming at comparing their performances.