Humphrey Sorensen

Stable cooperation in the N-player prisoner's dilemma: the importance of community structure

N-player prisoner dilemma games have been adopted and studied as a representation of many social dilemmas. They capture a larger class of social dilemmas than the traditional two-player prisoners dilemma. In N-player games, defection is the individually rational strategy and normally emerges as the dominant strategy in evolutionary simulations of agents playing the game. In this paper, we discuss the effect of a specific type of spatial constraint on a population of learning agents by placing agents on a graph structure which exhibits a community structure. We show that, by organising agents on a graph with a community structure, cooperation can exist despite the presence of defectors. Furthermore, we show that, by allowing agents learn from agents in neighbouring communities, cooperation can actually spread and become the dominant robust strategy. Moreover, we show that the spread of cooperation is robust to the introduction of noise into the system.

An intelligent agent for high-precision text filtering

We present here an overview of a research project aimed at reducing information overload for individual computer users. High-precision information filtering software has been developed to disseminate on–line electronic information. While the robustness and scalability of statistical approaches to information retrieval were a major influence on our design, we looked to the AI literature to supply the necessary techniques for the creation of an adaptive system. The system, called INFOrmer, is based on art intelligent agent approach and embodies machine learning, adaptation and relevance feedback techniques in its construction. A weighted graph representation is used for documents, and graph manipulation algorithms are used in the processing.

Investigations into user rating information and predictive accuracy in a collaborative filtering domain

The work described in this paper extracts user rating information from collaborative filtering datasets, and for each dataset uses a supervised machine learning approach to identify if there is an underlying relationship between rating information in the dataset and the expected accuracy of recommendations returned by the system. The underlying relationship is represented by decision tree rules. The rules can be used to indicate the predictive accuracy of the system for users of the system. Thus a user can know in advance of recommendation the level of accuracy to expect from the collaborative filtering system and may have more (or less) confidence in the recommendations produced. The experiment outlined in this paper aims to test the accuracy of the rules produced using three different datasets. Results show good accuracy can be found for all three datasets.

Co-evolution of strategies for an n-player dilemma

This work presents results on co-evolving classes of strategies for the n-player iterated prisoners dilemma (NIPD). We incorporate the notion of forgiveness in strategies and present experimental results which show that higher levels of cooperation and fitness are attainable when strategies are forgiving.

Information Foraging with a Proximity-Based Browsing Tool

This paper describes a user friendly, powerful information foraging tool. Document sets are presented through combinations of traditional ranked lists and 2-dimensional proximity-based visualisations, created by uniting graph-theoretic clustering and force-directed layout techniques, where article positions are determined by inter-document similarities. By using Gestalt principles and information encoding, the simple layout improves search efficiency by leveraging human cognitive strengths that have generally been under-utilised in commercial GUI development. In this paper, design and realisation of the layout technique are described in the context of an article browsing framework. Results of an indicative comparative laboratory study, which evaluates the client application – and in particular Graph-Theoretic Force-Directed (GTFD) visualisations against traditional search engine interfaces – are then presented. This study demonstrates the advantage of graphical presentations when browsing an article collection. Finally, potential improvements identified during the study are discussed, as are future directions for this approach to collection browsing

A constrained spreading activation approach to collaborative filtering

In this paper, we describe a collaborative filtering approach that aims to use features of users and items to better represent the problem space and to provide better recommendations to users. The goal of the work is to show that a graph-based representation of the problem domain, and a constrained spreading activation approach to effect retrieval, has as good, or better, performance than a traditional collaborative filtering approach using Pearson Correlation. However, in addition, the representation and approach proposed can be easily extended to incorporate additional information.

Evolving cultural learning parameters in an NK fitness landscape

Cultural learning allows individuals to acquire knowledge from others through non-genetic means. The effect of cultural learning on the evolution of artificial organisms has been the focus of much research. This paper examines the effects of cultural learning on the fitness and diversity of a population and, in addition, the effect of self-adaptive cultural learning parameters on the evolutionary process. The NK fitness landscape model is employed as the problem task and experiments employing populations endowed with both evolutionary and cultural learning are compared to those employing evolutionary learning alone. Our experiments measure the fitness and diversity of both populations and also track the values of two self-adaptive cultural parameters. Results show that the addition of cultural learning has a beneficial effect on the population in terms of fitness and diversity maintenance. Furthermore, analysis of the self-adaptive parameter values shows the relative quality of the cultural process throughout the experiment and highlights the benefits of self-adaptation over fixed parameter values.

Development of a rule-based finite capacity scheduling system

Abstract This paper discusses the use of the finite capacity planning model as a basis for a job-shop scheduling system. The suitability of this approach for constructing short term schedules with very short lead times is examined. This paper introduces a rule-based scheduling system whose operation, unlike previous monolithic schedulers, is based around distinct, though interlinked, processes. Scheduling herein is defined as the selection of a set of orders for manufacture, and the allocation of processing time to each on an acyclic network of processing resources. The objectives are, firstly, compliance with all customer deadlines and, secondly, the efficient utilisation of available machine time and prevailing machine set-ups.

Application of session analysis to search interface design

Evaluations of search features used in digital library environments are generally results centric, focussing on the outcome of an evaluation - for example, the number of relevant documents retrieved - rather than garnering an understanding of why that result was achieved. This paper explores how search feature development benefits from user-centered evaluation. By examining the application of an established web analytics technique, session analysis, to the development of search features and interfaces, it will be shown that designers can better understand how users conduct evaluation tasks. The feedback provided by this technique allows for clearer evaluation of an interface and admits iteratively evolving designs that are based on empirical data.

The effects of lifetime learning on the diversity and fitness of populations

Evolutionary learning refers to the process whereby a population of organisms evolves, or learns, by genetic means through a Darwinian process of iterated selection and reproduction of fit individuals. Hinton and Nowlan employed a genetic algorithm to study the effects of lifetime learning on the performance of genetic evolution [1]. Each agent in the model possesses a genome, comprised of a string of characters which can be one of 1, 0 or ?. Each agent is allowed a number of rounds of lifetime learning where for each ? in the genotype they ‘guess’ its value, assigning it either a 1 or a 0. Experimental results showed that, once learning was applied, the population converged on the problem solution, showing that individual learning is capable of guiding genetic evolution.