On the limitations of scalarisation for multi-objective reinforcement learning of Pareto fronts
- Authors: Vamplew, Peter , Yearwood, John , Dazeley, Richard , Berry, Adam
- Date: 2008
- Type: Text , Conference paper
- Relation: Paper presented at 21st Australasian Joint Conference on Artificial Intelligence, Auckland, New Zealand : 1st-5th December 2008 Vol. 5360, p. 372-378
- Full Text: false
- Description: Multiobjective reinforcement learning (MORL) extends RL to problems with multiple conflicting objectives. This paper argues for designing MORL systems to produce a set of solutions approximating the Pareto front, and shows that the common MORL technique of scalarisation has fundamental limitations when used to find Pareto-optimal policies. The work is supported by the presentation of three new MORL benchmarks with known Pareto fronts.
- Description: 2003006504
Online group deliberation for the elicitation of shared values to underpin decision making
- Authors: Feldman, Yishai , Kraft, Donald , Kuflik, Tsvi , Afshar, Faezeh , Stranieri, Andrew , Yearwood, John
- Date: 2009
- Type: Text , Conference paper
- Relation: Paper presented at 7th International Conference, NGITS 2009, Next generation information technologies and systems, Haifa, Israel : 16th-18th June 2009 Vol. 5831, p. 158-168
- Full Text: false
- Description: Values have been shown to underpin our attitudes, behaviour and motivate our decisions. Values do not exist in isolation but have meaning in relation to other values. However, values are not solely the purview of individuals as communities and organisations have core values implicit in their culture, policies and practices. Values for a group can be determined by a minority in power, derived by algorithmically merging values each group member holds, or set by deliberative consensus. The elicitation of values for the group by deliberation is likely to lead to widespread acceptance of values arrived at, however enticing individuals to engage in face to face discussion about values has been found to be very difficult. We present an online deliberative communication approach for the anonymous deliberation of values and claim that the framework has the elements required for the elicitation of shared values.
- Description: 2003007509
Opinion search in web logs
- Authors: Osman, Deanna , Yearwood, John
- Date: 2007
- Type: Text , Conference paper
- Relation: Paper presented at Eighteenth Australasian Database Conference, ADC 2007, Ballarat, Victoria : 29th January-2nd February 2007 p. 133-139
- Full Text:
- Description: Web logs(blogs) are a fast growing forum for people of all ages to express their feelings and opinions on topics of interest. The entries are often written in informal language without the structure found in newswire or published articles. One blog entry may contain many topics, these topics may express an opinion or a fact on a particular topic. This research is in contrast to work on opinion detection which has been carried out on more formally authored texts and on segments that are either whole documents or sentences. Whole web logs are divided into topics using a simple text segmentation approach. Similarity scores are used to distinguish where topic changers occur. The results are compared to human-evaluated topic changes and the most accurate algorithm is used in the remainder of the research. Words within each topic-block are allocated weightings depending on their opinion-bearing strength. Two approaches of using these weights, the sum and the maximum, are used to determine whether the topic-block is opinion-bearing or non-opinion-bearing. The opinion-bearing topic-blocks are rated by human evaluators as either opinion-bearing or non-opinion-bearing with precision of 67% for approach A and 70% for approach B. These results are compared with two approaches on published text to identify the difference between web logs and published articles.
- Description: 2003004895
Optimal rees matrix constructions for analysis of data
- Authors: Kelarev, Andrei , Yearwood, John , Zi, Lifang
- Date: 2012
- Type: Text , Journal article
- Relation: Journal of the Australian Mathematical Society Vol. 92, no. 3 (2012), p. 357-366
- Relation: http://purl.org/au-research/grants/arc/LP0990908
- Relation: http://purl.org/au-research/grants/arc/DP0211866
- Full Text:
- Reviewed:
- Description: Abstract We introduce a new construction involving Rees matrix semigroups and max-plus algebras that is very convenient for generating sets of centroids. We describe completely all optimal sets of centroids for all Rees matrix semigroups without any restrictions on the sandwich matrices. © 2013 Australian Mathematical Publishing Association Inc.
- Description: 2003010862
Optimization and matrix constructions for classification of data
- Authors: Kelarev, Andrei , Yearwood, John , Vamplew, Peter , Abawajy, Jemal , Chowdhury, Morshed
- Date: 2011
- Type: Journal article
- Relation: New Zealand Journal of Mathematics Vol. 41, no. 2011 (2011), p. 65-73
- Full Text:
- Reviewed:
- Description: Max-plus alegbras and more general semirings have many useful applications and have been actively investigated. On the other hand, structural matrix rings are also well known and have been considered by many authors. The main theorem of this article completely describes all optimal ideas in the more general structural matrix semirings. Originally, our investigation of these ideals was motivated by applications in data mining for the design of multiple classification systems combining several individual classifiers.
Optimization methods and the k-committees algorithm for clustering of sequence data
- Authors: Yearwood, John , Bagirov, Adil , Kelarev, Andrei
- Date: 2009
- Type: Text , Journal article
- Relation: Applied and Computational Mathematics Vol. 8, no. 1 (2009), p. 92-101
- Relation: http://purl.org/au-research/grants/arc/DP0211866
- Relation: http://purl.org/au-research/grants/arc/DP0666061
- Full Text: false
- Description: The present paper is devoted to new algorithms for unsupervised clustering based on the optimization approaches due to [2], [3] and [4]. We consider a novel situation, where the datasets consist of nucleotide or protein sequences and rather sophisticated biologically significant alignment scores have to be used as a measure of distance. Sequences of this kind cannot be regarded as points in a finite dimensional space. Besides, the alignment scores do not satisfy properties of Minkowski metrics. Nevertheless the optimization approaches have made it possible to introduce a new k-committees algorithm and compare its performance with previous algorithms for two datasets. Our experimental results show that the k-committees algorithms achieves intermediate accuracy for a dataset of ITS sequences, and it can perform better than the discrete k-means and Nearest Neighbour algorithms for certain datasets. All three algorithms achieve good agreement with clusters published in the biological literature before and can be used to obtain biologically significant clusterings.
Optimization of classifiers for data mining based on combinatorial semigroups
- Authors: Kelarev, Andrei , Yearwood, John , Watters, Paul
- Date: 2011
- Type: Text , Journal article
- Relation: Semigroup Forum Vol. 82, no. 2 (2011), p. 1-10
- Full Text:
- Reviewed:
- Description: The aim of the present article is to obtain a theoretical result essential for applications of combinatorial semigroups for the design of multiple classification systems in data mining. We consider a novel construction of multiple classification systems, or classifiers, combining several binary classifiers. The construction is based on combinatorial Rees matrix semigroups without any restrictions on the sandwich-matrix. Our main theorem gives a complete description of all optimal classifiers in this novel construction. © 2011 Springer Science+Business Media, LLC.
Optimization of feed forward MLPs using the discrete gradient method
- Authors: Bagirov, Adil , Yearwood, John , Ghosh, Ranadhir
- Date: 2004
- Type: Text , Conference paper
- Relation: Paper presented at CIMCA 2004: International Conference on Computational Intelligence for Modelling, Control & Automation, Gold Coast, Queensland : 12th July, 2004
- Full Text: false
- Reviewed:
- Description: E1
- Description: 2003000845
Optimization of matrix semirings for classification systems
- Authors: Gao, David , Kelarev, Andrei , Yearwood, John
- Date: 2011
- Type: Text , Journal article
- Relation: Bulletin of the Australian Mathematical Society Vol. 84, no. 3 (2011), p. 492-503
- Full Text:
- Reviewed:
- Description: The max-plus algebra is well known and has useful applications in the investigation of discrete event systems and affine equations. Structural matrix rings have been considered by many authors too. This article introduces more general structural matrix semirings, which include all matrix semirings over the max-plus algebra. We investigate properties of ideals in this construction motivated by applications to the design of centroid-based classification systems, or classifiers, as well as multiple classifiers combining several initial classifiers. The first main theorem of this paper shows that structural matrix semirings possess convenient visible generating sets for ideals. Our second main theorem uses two special sets to determine the weights of all ideals and describe all matrix ideals with the largest possible weight, which are optimal for the design of classification systems. © Copyright Australian Mathematical Publishing Association Inc. 2011.
- Description: 2003009498
Optimization of multiple classifiers in data mining based on string rewriting systems
- Authors: Dazeley, Richard , Kelarev, Andrei , Yearwood, John , Mammadov, Musa
- Date: 2009
- Type: Text , Journal article
- Relation: Asian-European Journal of Mathematics Vol. 2, no. 1 (2009), p. 41-56
- Relation: https://purl.org/au-research/grants/arc/DP0211866
- Relation: https://purl.org/au-research/grants/arc/LP0669752
- Full Text:
- Description: Optimization of multiple classifiers is an important problem in data mining. We introduce additional structure on the class sets of the classifiers using string rewriting systems with a convenient matrix representation. The aim of the present paper is to develop an efficient algorithm for the optimization of the number of errors of individual classifiers, which can be corrected by these multiple classifiers.
Parallel selection of multi-category features for online handwritten character recognition
- Authors: Ghosh, Ranadhir , Ghosh, Moumita , Yearwood, John
- Date: 2005
- Type: Text , Conference proceedings
- Full Text: false
- Description: Online handwritten recognition is gaining more interest due to the increasing popularity of hand-held computers, digital notebooks and advanced cellular phones. The large number of writing styles and the variability between them makes the handwriting recognition problem a very challenging area for researchers. Many previous efforts have utilized many different approaches for recognition in online handwriting using various ANN classifier-modeling techniques. Different types of feature extraction techniques have also been used. It has been observed that, beyond a certain point, the inclusion of additional features leads to a worse rather than better performance. Moreover, the choice of features to represent the patterns affects several aspects of pattern recognition problems such as accuracy, required learning time and a necessary number of samples. A common problem with the multi-category feature classification is the conflict between the categories. None of the feasible solutions allow simultaneous optimal solution for all categories. In order to find an optimal solution the search space can be divided based on an individual category in each sub region and finally merging them through decision spport system. In this paper we propose a canonical GA based modular feature selection approach combined with standard MLP for multi category feature selection in online handwriting recognition.
Performance evaluation of multi-tier ensemble classifiers for phishing websites
- Authors: Abawajy, Jemal , Beliakov, Gleb , Kelarev, Andrei , Yearwood, John
- Date: 2012
- Type: Text , Conference proceedings
- Full Text:
- Description: This article is devoted to large multi-tier ensemble classifiers generated as ensembles of ensembles and applied to phishing websites. Our new ensemble construction is a special case of the general and productive multi-tier approach well known in information security. Many efficient multi-tier classifiers have been considered in the literature. Our new contribution is in generating new large systems as ensembles of ensembles by linking a top-tier ensemble to another middletier ensemble instead of a base classifier so that the toptier ensemble can generate the whole system. This automatic generation capability includes many large ensemble classifiers in two tiers simultaneously and automatically combines them into one hierarchical unified system so that one ensemble is an integral part of another one. This new construction makes it easy to set up and run such large systems. The present article concentrates on the investigation of performance of these new multi-tier ensembles for the example of detection of phishing websites. We carried out systematic experiments evaluating several essential ensemble techniques as well as more recent approaches and studying their performance as parts of multi-level ensembles with three tiers. The results presented here demonstrate that new three-tier ensemble classifiers performed better than the base classifiers and standard ensembles included in the system. This example of application to the classification of phishing websites shows that the new method of combining diverse ensemble techniques into a unified hierarchical three-tier ensemble can be applied to increase the performance of classifiers in situations where data can be processed on a large computer.
Performance evaluation of multivariate non-normal process using metaheuristic approaches
- Authors: Ahmad, S. , Abdollahian, Mali , Bhatti, M.I. , Huda, Shamsul , Yearwood, John
- Date: 2014
- Type: Text , Journal article
- Relation: Journal of Applied Statistical Science Vol. 20, no. 3 (2014), p. 299-315
- Full Text: false
- Reviewed:
- Description: Multivariate process performance indices generally rely on the assumption that the process follow normal distribution but in practice its non-normal with correlated characteristics patterns. This paper proposes two metaheuristic-based approaches to fit Burr distribution to such data; a single candidate model based approach using a Simulated Annealing (SA) technique and a population based approach using a constraint-based Evolutionary Alogorithn (EA). The fitted Burr distribution is then used to estimate the proportion of Non-conforming (PNC) which is then used to fit an appropiate Burr distribution to individual Geometric distance variables. Empirical performance of the proposed methods have been evaluated on real industrial data set using PNC criterion. Experimental results demonstrate that the new approach perform well than the existing.
Predicting Australian stock market index using neural networks exploiting dynamical swings and intermarket influences
- Authors: Pan, Heping , Tilakaratne, Chandima , Yearwood, John
- Date: 2005
- Type: Text , Journal article
- Relation: Journal of Research and Practice in Information Technology Vol. 37, no. 1 (2005), p. 43-55
- Full Text:
- Reviewed:
- Description: This paper presents a computational approach for predicting the Australian stock market index AORD using multi-layer feed-forward neural networks front the time series data of AORD and various interrelated markets. This effort aims to discover an effective neural network, or a set of adaptive neural networks for this prediction purpose, which can exploit or model various dynamical swings and inter-market influences discovered from professional technical analysis and quantitative analysis. Within a limited range defined by our empirical knowledge, three aspects of effectiveness on data selection are considered: effective inputs from the target market (AORD) itself, a sufficient set of interrelated markets,. and effective inputs from the interrelated markets. Two traditional dimensions of the neural network architecture are also considered: the optimal number of hidden layers, and the optimal number of hidden neurons for each hidden layer. Three important results were obtained: A 6-day cycle was discovered in the Australian stock market during the studied period; the time signature used as additional inputs provides useful information; and a basic neural network using six daily returns of AORD and one daily, returns of SP500 plus the day of the week as inputs exhibits up to 80% directional prediction correctness.
- Description: C1
- Description: 2003001440
Predicting the Australian stock market index using neural networks and exploiting dynamical swings and intermarket influences
- Authors: Pan, Heping , Tilakaratne, Chandima , Yearwood, John
- Date: 2003
- Type: Text , Conference paper
- Relation: Paper presented at AI 2003: Advances in Artificial Intelligence - the 16th Australian Conference on AI, Perth : 3rd December, 2003
- Full Text: false
- Reviewed:
- Description: This paper presents a computational approach for predicting the Australian stock market index - AORD using multi-layer feed-forward neural networks from the time series data of AORD and various interrelated markets. This effort aims to discover an optimal neural network or a set of adaptive neural networks for this prediction purpose, which can exploit or model various dynamical swings and intermarket influences discovered from professional technical analysis and quantitative analysis. Four dimensions for optimality on data selection are considered: the optimal inputs from the target market (AORD) itself, the optimal set of interrelated markets, the optimal inputs from the optimal interrelated markets, and the optimal outputs. Two traditional dimensions of the neural network architecture are also considered: the optimal number of hidden layers, and the optimal number of hidden neurons for each hidden layer. Three important results were obtained: A 6-day cycle was discovered in the Australian stock market; the time signature used as additional inputs provides useful information; and a minimal neural network using 6 daily returns of AORD and 1 daily returns of SP500 plus the day of the week as inputs exhibits up to 80% directional prediction correctness.
- Description: E1
- Description: 2003000374
Profiling phishing activity based on hyperlinks extracted from phishing emails
- Authors: Yearwood, John , Mammadov, Musa , Webb, Dean
- Date: 2012
- Type: Text , Journal article
- Relation: Social Network Analysis and Mining Vol. 2, no. 1 (2012), p. 5-16
- Full Text: false
- Reviewed:
- Description: Phishing activity has recently been focused on social networking sites as a more effective way of exploiting not only the technology but also the trust that may exist between members in a social network. In this paper, a novel method for profiling phishing activity from an analysis of phishing emails is proposed. Profiling is useful in determining the activity of an individual or a particular group of phishers. Work in the area of phishing is usually aimed at detection of phishing emails. In this paper, we concentrate on profiling as distinct from detection of phishing emails. We formulate the profiling problem as a multi-label classification problem using the hyperlinks in the phishing emails as features and structural properties of emails along with whois (i.e. DNS) information on hyperlinks as profile classes. Further, we generate profiles based on the classifier predictions. Thus, classes become elements of profiles. We employ a boosting algorithm (AdaBoost) as well as SVM to generate multi-label class predictions on three different datasets created from hyperlink information in phishing emails. These predictions are further utilized to generate complete profiles of these emails. Results show that profiling can be done with quite high accuracy using hyperlink information.
Profiling phishing emails based on hyperlink information
- Authors: Yearwood, John , Mammadov, Musa , Banerjee, Arunava
- Date: 2010
- Type: Text , Conference paper
- Relation: Paper presented at 2010 International Conference on Advances in Social Network Analysis and Mining, ASONAM 2010, Odense : 9th-11th August 2010 p. 120-127
- Full Text:
- Description: In this paper, a novel method for profiling phishing activity from an analysis of phishing emails is proposed. Profiling is useful in determining the activity of an individual or a particular group of phishers. Work in the area of phishing is usually aimed at detection of phishing emails. In this paper, we concentrate on profiling as distinct from detection of phishing emails. We formulate the profiling problem as a multi-label classification problem using the hyperlinks in the phishing emails as features and structural properties of emails along with whois (i.e.DNS) information on hyperlinks as profile classes. Further, we generate profiles based on classifier predictions. Thus, classes become elements of profiles. We employ a boosting algorithm (AdaBoost) as well as SVM to generate multi-label class predictions on three different datasets created from hyperlink information in phishing emails. These predictions are further utilized to generate complete profiles of these emails. Results show that profiling can be done with quite high accuracy using hyperlink information. © 2010 Crown Copyright.
Real-time detection of children's skin on social networking sites using Markov random field modelling
- Authors: Islam, Mofakharul , Watters, Paul , Yearwood, John
- Date: 2011
- Type: Text , Journal article
- Relation: Information Security Technical Report Vol. 16, no. 2 (2011), p. 51-58
- Full Text: false
- Reviewed:
- Description: Social networking sites are increasingly being used as the source for paedophiles to search for, download and exchange child exploitation images. Law Enforcement Agencies (LEAs) around the world face a difficult challenge to combat technologically-savvy paedophiles. In this paper, we propose a framework for detecting images containing children's pictures in different poses, with the ultimate view of identifying and classifying images as corresponding to the COPINE scale. To achieve the goal of automatic detection, we present a novel stochastic vision model based on a Markov Random Fields (MRF) prior, which will employ a skin model and human affine-invariant geometric descriptor to detect and identify skin regions containing pornographic contexts. © 2011 Published by Elsevier Ltd.
Rees matrix constructions for clustering of data
- Authors: Kelarev, Andrei , Watters, Paul , Yearwood, John
- Date: 2009
- Type: Journal article
- Relation: Journal of the Australian Mathematical Society Vol. 87, no. 3 (2009), p. 377-393
- Relation: http://purl.org/au-research/grants/arc/DP0211866
- Full Text:
- Reviewed:
- Description: This paper continues the investigation of semigroup constructions motivated by applications in data mining. We give a complete description of the error-correcting capabilities of a large family of clusterers based on Rees matrix semigroups well known in semigroup theory. This result strengthens and complements previous formulas recently obtained in the literature. Examples show that our theorems do not generalize to other classes of semigroups.
Reinforcement learning approach to AIBO robot's decision making process in Robosoccer's goal keeper problem
- Authors: Mukherjee, Subhasis , Yearwood, John , Vamplew, Peter , Huda, Shamsul
- Date: 2011
- Type: Text , Conference proceedings
- Full Text: false
- Description: Robocup is a popular test bed for AI programs around the world. Robosoccer is one of the two major parts of Robocup, in which AIBO entertainment robots take part in the middle sized soccer event. The three key challenges that robots need to face in this event are manoeuvrability, image recognition and decision making skills. This paper focuses on the decision making problem in Robosoccer - The goal keeper problem. We investigate whether reinforcement learning (RL) as a form of semi-supervised learning can effectively contribute to the goal keeper's decision making process when penalty shot and two attacker problem are considered. Currently, the decision making process in Robosoccer is carried out using rule-base system. RL also is used for quadruped locomotion and navigation purpose in Robosoccer using AIBO. In this paper, we propose a reinforcement learning based approach that uses a dynamic state-action mapping using back propagation of reward and space quantized Q-learning (SQQL) for the choice of high level functions in order to save the goal. The novelty of our approach is that the agent learns while playing and can take independent decision which overcomes the limitations of rule-base system due to fixed and limited predefined decision rules. Performance of the proposed method has been verified against the bench mark data set made with Upenn'03 code logic. It was found that the efficiency of our SQQL approach in goalkeeping was better than the rule based approach. The SQQL develops a semi-supervised learning process over the rule-base system's input-output mapping process, given in the Upenn'03 code. © 2011 IEEE.