arXiv subjects

Tshilidzi Marwala

Publications and source records attributed to Tshilidzi Marwala.

82 records · Page 5Linked to original sources

arXiv2007

Control of Complex Systems Using Bayesian Networks and Genetic Algorithm

A method based on Bayesian neural networks and genetic algorithm is proposed to control the fermentation process. The relationship between input and output variables is modelled using Bayesian neural network that is trained using hybrid Monte Carlo method. A feedback loop based on genetic algorithm is used to change input variables so that the output variables are as close to the desired target as possible without the loss of confidence level on the prediction that the neural network gives. The proposed procedure is found to reduce the distance between the desired target and measured outputs significantly.

cs.CE↗

arXiv2007

Learning to Bluff

The act of bluffing confounds game designers to this day. The very nature of bluffing is even open for debate, adding further complication to the process of creating intelligent virtual players that can bluff, and hence play, realistically. Through the use of intelligent, learning agents, and carefully designed agent outlooks, an agent can in fact learn to predict its opponents reactions based not only on its own cards, but on the actions of those around it. With this wider scope of understanding, an agent can in learn to bluff its opponents, with the action representing not an illogical action, as bluffing is often viewed, but rather as an act of maximising returns through an effective statistical optimisation. By using a tee dee lambda learning algorithm to continuously adapt neural network agent intelligence, agents have been shown to be able to learn to bluff without outside prompting, and even to learn to call each others bluffs in free, competitive play.

cs.AI↗

arXiv2007

An Independent Evaluation of Subspace Face Recognition Algorithms

This paper explores a comparative study of both the linear and kernel implementations of three of the most popular Appearance-based Face Recognition projection classes, these being the methodologies of Principal Component Analysis, Linear Discriminant Analysis and Independent Component Analysis. The experimental procedure provides a platform of equal working conditions and examines the ten algorithms in the categories of expression, illumination, occlusion and temporal delay. The results are then evaluated based on a sequential combination of assessment tools that facilitate both intuitive and statistical decisiveness among the intra and interclass comparisons. The best categorical algorithms are then incorporated into a hybrid methodology, where the advantageous effects of fusion strategies are considered.

cs.CV↗

arXiv2007

Artificial Neural Networks and Support Vector Machines for Water Demand Time Series Forecasting

Water plays a pivotal role in many physical processes, and most importantly in sustaining human life, animal life and plant life. Water supply entities therefore have the responsibility to supply clean and safe water at the rate required by the consumer. It is therefore necessary to implement mechanisms and systems that can be employed to predict both short-term and long-term water demands. The increasingly growing field of computational intelligence techniques has been proposed as an efficient tool in the modelling of dynamic phenomena. The primary objective of this paper is to compare the efficiency of two computational intelligence techniques in water demand forecasting. The techniques under comparison are the Artificial Neural Networks (ANNs) and the Support Vector Machines (SVMs). In this study it was observed that the ANNs perform better than the SVMs. This performance is measured against the generalisation ability of the two.

cs.AI↗

arXiv2007

Fault Classification in Cylinders Using Multilayer Perceptrons, Support Vector Machines and Guassian Mixture Models

Gaussian mixture models (GMM) and support vector machines (SVM) are introduced to classify faults in a population of cylindrical shells. The proposed procedures are tested on a population of 20 cylindrical shells and their performance is compared to the procedure, which uses multi-layer perceptrons (MLP). The modal properties extracted from vibration data are used to train the GMM, SVM and MLP. It is observed that the GMM produces 98%, SVM produces 94% classification accuracy while the MLP produces 88% classification rates.

cs.AI↗

arXiv2007

Using Images to create a Hierarchical Grid Spatial Index

This paper presents a hybrid approach to spatial indexing of two dimensional data. It sheds new light on the age old problem by thinking of the traditional algorithms as working with images. Inspiration is drawn from an analogous situation that is found in machine and human vision. Image processing techniques are used to assist in the spatial indexing of the data. A fixed grid approach is used and bins with too many records are sub-divided hierarchically. Search queries are pre-computed for bins that do not contain any data records. This has the effect of dividing the search space up into non rectangular regions which are based on the spatial properties of the data. The bucketing quad tree can be considered as an image with a resolution of two by two for each layer. The results show that this method performs better than the quad tree if there are more divisions per layer. This confirms our suspicions that the algorithm works better if it gets to look at the data with higher resolution images. An elegant class structure is developed where the implementation of concrete spatial indexes for a particular data type merely relies on rendering the data onto an image.

cs.DS↗

arXiv2007

An Integrated Human-Computer System for Controlling Interstate Disputes

In this paper we develop a scientific approach to control inter-country conflict. This system makes use of a neural network and a feedback control approach. It was found that by controlling the four controllable inputs: Democracy, Dependency, Allies and Capability simultaneously, all the predicted dispute outcomes could be avoided. Furthermore, it was observed that controlling a single input Dependency or Capability also avoids all the predicted conflicts. When the influence of each input variable on conflict is assessed, Dependency, Capability, and Democracy emerge as key variables that influence conflict.

stat.AP↗

arXiv2007

Missing Data: A Comparison of Neural Network and Expectation Maximisation Techniques

The estimation of missing input vector elements in real time processing applications requires a system that possesses the knowledge of certain characteristics such as correlations between variables, which are inherent in the input space. Computational intelligence techniques and maximum likelihood techniques do possess such characteristics and as a result are important for imputation of missing data. This paper compares two approaches to the problem of missing data estimation. The first technique is based on the current state of the art approach to this problem, that being the use of Maximum Likelihood (ML) and Expectation Maximisation (EM. The second approach is the use of a system based on auto-associative neural networks and the Genetic Algorithm as discussed by Adbella and Marwala3. The estimation ability of both of these techniques is compared, based on three datasets and conclusions are made.

stat.AP↗

arXiv2007

Rough Sets Computations to Impute Missing Data

Many techniques for handling missing data have been proposed in the literature. Most of these techniques are overly complex. This paper explores an imputation technique based on rough set computations. In this paper, characteristic relations are introduced to describe incompletely specified decision tables.It is shown that the basic rough set idea of lower and upper approximations for incompletely specified decision tables may be defined in a variety of different ways. Empirical results obtained using real data are given and they provide a valuable and promising insight to the problem of missing data. Missing data were predicted with an accuracy of up to 99%.

cs.CV↗

arXiv2007

Bayesian approach to rough set

This paper proposes an approach to training rough set models using Bayesian framework trained using Markov Chain Monte Carlo (MCMC) method. The prior probabilities are constructed from the prior knowledge that good rough set models have fewer rules. Markov Chain Monte Carlo sampling is conducted through sampling in the rough set granule space and Metropolis algorithm is used as an acceptance criteria. The proposed method is tested to estimate the risk of HIV given demographic data. The results obtained shows that the proposed approach is able to achieve an average accuracy of 58% with the accuracy varying up to 66%. In addition the Bayesian rough set give the probabilities of the estimated HIV status as well as the linguistic rules describing how the demographic parameters drive the risk of HIV.

cs.AI↗