Robert J. Cox

A review of linden scripting language and its role in second life

The Second Life virtual world (SL) created by Linden Lab (LL) provides a rich three-dimensional environment, allowing the residents of this virtual community to create and trade unique content. Linden Scripting Language (LSL) is used to enhance the SL experience by providing a programming language capability for created objects. Primitives (prims) are the atomic objects from which complex objects are built; they can be scripted using LSL In this paper we specifically look at how LSL affects the SL experience, its strengths and weaknesses, and propose enhancements, also seeking to find where it has been innovative or unusual. One of the biggest problems in SL is lag – the way the experience slows under load. LSL scripts contribute considerably to lag and we look at how the design of the language attempts to assist in keeping the SL experience enjoyable.

Use of Artificial Neural Networks in the Prediction of Kidney Transplant Outcomes

Traditionally researchers have used statistical methods to predict medical outcomes. However, statistical techniques do not provide sufficient in-formation for solving problems of high complexity. Recently more attention has turned to a variety of artificial intelligence modeling techniques such as Artificial Neural Networks (ANNs), Case Based Reasoning (CBR) and Rule Induction (RI). In this study we sought to use ANN to predict renal transplantation outcomes. Our results showed that although this was possible, the positive predictive power of the trained ANN was low, indicating a need for improvement if this approach is to be useful clinically. We also highlight potential problems that may arise when using incomplete clinical datasets for ANN train-ing including the danger of pre-processing data in such a way that misleading high predictive value is obtained.

An Investigation into the Effect of Ensemble Size and Voting Threshold on the Accuracy of Neural Network Ensembles

If voting is used by an ensemble to classify data, some data points may not be classified, but a higher proportion of those which are classified are classified correctly. This trade off is affected by ensemble size and voting threshold. This paper investigates the effect of ensemble size on the proportions of decisions made and correct decisions. It does this for majority voting and consensus voting on ensembles of neural network classifiers constructed using bagging. It also models the relationships in order to estimate the asymptotic values as the ensemble size increases.

Building content in second life – issues facing content creators and residents

The advent of virtual communities in Massively Multiplayer Online Roleplay Games (MMORPGs) is a relatively recent phenomenon. One such virtual community, the Second Life world, allows its residents to create unique content, such as clothes, hair, buildings, furniture; even vehicles. The variety of possibilities is encompassed by the phrase, “Your World, Your Imagination” which features prominently on the Linden Lab web site. Content authors use tools provided by Linden Lab to create items for their own use, or to be given away or sold to other players. We examine the tools provided to produce content within Second Life, concentrating on issues caused by limitations in the tools, and proposing solutions to some of the more vexing problems caused by these limitations.

Accuracy of neural network classifiers as a property of the size of the data set

It is well-known that the accuracy of a neural network classifier increases as the number of data points in the training set increases. A previous researcher has proposed a general mathematical model that explains the relationship between training sample size and predictive power. We examine this model using artificially generated data sets containing varying numbers of data points and some real world data sets. We find the model works well when large numbers of data points are available for training, but presents practical difficulties when the amount of available data is small and the data set is difficult to classify.

A Study of the Radial Basis Function Neural Network Classifiers Using Known Data of Varying Accuracy and Complexity

Neural networks are increasingly used in a wide variety of applications such as speech recognition, diagnostic prediction, income prediction and credit screening. This paper empirically compares the performance of Radial Basis Function (RBF) and Multilayer Perceptron (MLP) neural networks using artificially generated data sets, enabling us to accurately chart the effectiveness of each network type and to provide some guidance to practitioners as to which type of network to use with their data. We find that when the discriminator is simple, RBF and MLP network performances are similar; when the number of data points is relatively small the MLP outperforms the RBF; when the discriminator is complex the RBF outperforms the MLP; and when the data has an unrelated input and the underlying discriminator is simple, the MLP outperforms the RBF.

A computationally efficient fuzzy logic parameterisation system for computer games

Linguistic fuzzy expert systems provide useful tools for the implementation of Artificial Intelligence (AI) systems for computer games. However, in games where a large number of fuzzy agents are needed, the computational needs of the fuzzy expert system inclines designers to abandon this promising technique in favour of non-fuzzy AI techniques with a lower computational overhead. In this paper we investigated a parameterisation of fuzzy sets with the goal of finding fuzzy systems that have lower computational needs but still have sufficient accuracy for use in the domain of computer games. We developed a system we call short-cut fuzzy logic that has low computational needs and seems to have adequate accuracy for the games domain.

Using artificial neural network ensembles to extract data content from noisy data

We have developed a technique to extract points that contain information from a sea of noisy data using an ensemble of Artificial Neural Networks. The technique is relatively simple to use and by using artificial data sets we demonstrate that it can extract a subset of the data that in effect has a higher signal to noise ratio than the original data. We assert that this technique is of practical use in the area of classification, although it does appear to lose points, particularly near the discriminator.