GYAN: A methodology for rule extraction from artificial neural networks

Artificial neural network (ANN) learning methods provide a robust and non-linear approach to approximating the target function for many classification, regression and clustering problems. ANNs have demonstrated good predictive performance in a wide variety of practical problems. However, there are strong arguments as to why ANNs are not sufficient for the general representation of knowledge. The arguments are the poor comprehensibility of the learned ANN, and the inability to represent explanation structures. The overall objective of this thesis is to address these issues by: (1) explanation of the decision process in ANNs in the form of symbolic rules (predicate rules with variables); and (2) provision of explanatory capability by mapping the general conceptual knowledge that is learned by the neural networks into a knowledge base to be used in a rule-based reasoning system. A multi-stage methodology GYAN is developed and evaluated for the task of extracting knowledge from the trained ANNs. The extracted knowledge is represented in the form of restricted first-order logic rules, and subsequently allows user interaction by interfacing with a knowledge based reasoner. The performance of GYAN is demonstrated using a number of real world and artificial data sets. The empirical results demonstrate that: (1) an equivalent symbolic interpretation is derived describing the overall behaviour of the ANN with high accuracy and fidelity, and (2) a concise explanation is given (in terms of rules, facts and predicates activated in a reasoning episode) as to why a particular instance is being classified into a certain category.

An evaluation of IS-impact utility and intuitiveness

This study is conducted within the IS-Impact Research Track at Queensland University of Technology (QUT). The goal of the IS-Impact Track is, "to develop the most widely employed model for benchmarking information systems in organizations for the joint benefit of both research and practice" (Gable et al, 2006). IS-Impact is defined as "a measure at a point in time, of the stream of net benefits from the IS [Information System], to date and anticipated, as perceived by all key-user-groups" (Gable Sedera and Chan, 2008). Track efforts have yielded the bicameral IS-Impact measurement model; the "impact" half includes Organizational-Impact and Individual-Impact dimensions; the "quality" half includes System-Quality and Information-Quality dimensions. The IS-Impact model, by design, is intended to be robust, simple and generalisable, to yield results that are comparable across time, stakeholders, different systems and system contexts. The model and measurement approach employs perceptual measures and an instrument that is relevant to key stakeholder groups, thereby enabling the combination or comparison of stakeholder perspectives. Such a validated and widely accepted IS-Impact measurement model has both academic and practical value. It facilitates systematic operationalisation of a main dependent variable in research (IS-Impact), which can also serve as an important independent variable. For IS management practice it provides a means to benchmark and track the performance of information systems in use. From examination of the literature, the study proposes that IS-Impact is an Analytic Theory. Gregor (2006) defines Analytic Theory simply as theory that ‘says what is’, base theory that is foundational to all other types of theory. The overarching research question thus is "Does IS-Impact positively manifest the attributes of Analytic Theory?" In order to address this question, we must first answer the question "What are the attributes of Analytic Theory?" The study identifies the main attributes of analytic theory as: (1) Completeness, (2) Mutual Exclusivity, (3) Parsimony, (4) Appropriate Hierarchy, (5) Utility, and (6) Intuitiveness. The value of empirical research in Information Systems is often assessed along the two main dimensions - rigor and relevance. Those Analytic Theory attributes associated with the ‘rigor’ of the IS-Impact model; namely, completeness, mutual exclusivity, parsimony and appropriate hierarchy, have been addressed in prior research (e.g. Gable et al, 2008). Though common tests of rigor are widely accepted and relatively uniformly applied (particularly in relation to positivist, quantitative research), attention to relevance has seldom been given the same systematic attention. This study assumes a mainly practice perspective, and emphasises the methodical evaluation of the Analytic Theory ‘relevance’ attributes represented by the Utility and Intuitiveness of the IS-Impact model. Thus, related research questions are: "Is the IS-Impact model intuitive to practitioners?" and "Is the IS-Impact model useful to practitioners?" March and Smith (1995), identify four outputs of Design Science: constructs, models, methods and instantiations (Design Science research may involve one or more of these). IS-Impact can be viewed as a design science model, composed of Design Science constructs (the four IS-Impact dimensions and the two model halves), and instantiations in the form of management information (IS-Impact data organised and presented for management decision making). In addition to methodically evaluating the Utility and Intuitiveness of the IS-Impact model and its constituent constructs, the study aims to also evaluate the derived management information. Thus, further research questions are: "Is the IS-Impact derived management information intuitive to practitioners?" and "Is the IS-Impact derived management information useful to practitioners? The study employs a longitudinal design entailing three surveys over 4 years (the 1st involving secondary data) of the Oracle-Financials application at QUT, interspersed with focus groups involving senior financial managers. The study too entails a survey of Financials at four other Australian Universities. The three focus groups respectively emphasise: (1) the IS-Impact model, (2) the 2nd survey at QUT (descriptive), and (3) comparison across surveys within QUT, and between QUT and the group of Universities. Aligned with the track goal of producing IS-Impact scores that are highly comparable, the study also addresses the more specific utility-related questions, "Is IS-Impact derived management information a useful comparator across time?" and "Is IS-Impact derived management information a useful comparator across universities?" The main contribution of the study is evidence of the utility and intuitiveness of IS-Impact to practice, thereby further substantiating the practical value of the IS-Impact approach; and also thereby motivating continuing and further research on the validity of IS-Impact, and research employing the ISImpact constructs in descriptive, predictive and explanatory studies. The study also has value methodologically as an example of relatively rigorous attention to relevance. A further key contribution is the clarification and instantiation of the full set of analytic theory attributes.

Hierarchical models for 2D presence/absence data having ambiguous zeroes: With a biogeographical case study on dingo behaviour

This dissertation is primarily an applied statistical modelling investigation, motivated by a case study comprising real data and real questions. Theoretical questions on modelling and computation of normalization constants arose from pursuit of these data analytic questions. The essence of the thesis can be described as follows. Consider binary data observed on a two-dimensional lattice. A common problem with such data is the ambiguity of zeroes recorded. These may represent zero response given some threshold (presence) or that the threshold has not been triggered (absence). Suppose that the researcher wishes to estimate the effects of covariates on the binary responses, whilst taking into account underlying spatial variation, which is itself of some interest. This situation arises in many contexts and the dingo, cypress and toad case studies described in the motivation chapter are examples of this. Two main approaches to modelling and inference are investigated in this thesis. The first is frequentist and based on generalized linear models, with spatial variation modelled by using a block structure or by smoothing the residuals spatially. The EM algorithm can be used to obtain point estimates, coupled with bootstrapping or asymptotic MLE estimates for standard errors. The second approach is Bayesian and based on a three- or four-tier hierarchical model, comprising a logistic regression with covariates for the data layer, a binary Markov Random field (MRF) for the underlying spatial process, and suitable priors for parameters in these main models. The three-parameter autologistic model is a particular MRF of interest. Markov chain Monte Carlo (MCMC) methods comprising hybrid Metropolis/Gibbs samplers is suitable for computation in this situation. Model performance can be gauged by MCMC diagnostics. Model choice can be assessed by incorporating another tier in the modelling hierarchy. This requires evaluation of a normalization constant, a notoriously difficult problem. Difficulty with estimating the normalization constant for the MRF can be overcome by using a path integral approach, although this is a highly computationally intensive method. Different methods of estimating ratios of normalization constants (N Cs) are investigated, including importance sampling Monte Carlo (ISMC), dependent Monte Carlo based on MCMC simulations (MCMC), and reverse logistic regression (RLR). I develop an idea present though not fully developed in the literature, and propose the Integrated mean canonical statistic (IMCS) method for estimating log NC ratios for binary MRFs. The IMCS method falls within the framework of the newly identified path sampling methods of Gelman & Meng (1998) and outperforms ISMC, MCMC and RLR. It also does not rely on simplifying assumptions, such as ignoring spatio-temporal dependence in the process. A thorough investigation is made of the application of IMCS to the three-parameter Autologistic model. This work introduces background computations required for the full implementation of the four-tier model in Chapter 7. Two different extensions of the three-tier model to a four-tier version are investigated. The first extension incorporates temporal dependence in the underlying spatio-temporal process. The second extensions allows the successes and failures in the data layer to depend on time. The MCMC computational method is extended to incorporate the extra layer. A major contribution of the thesis is the development of a fully Bayesian approach to inference for these hierarchical models for the first time. Note: The author of this thesis has agreed to make it open access but invites people downloading the thesis to send her an email via the 'Contact Author' function.

On the data consumption benefits of accepting increased uncertainty

In the context of learning paradigms of identification in the limit, we address the question: why is uncertainty sometimes desirable? We use mind change bounds on the output hypotheses as a measure of uncertainty, and interpret ‘desirable’ as reduction in data memorization, also defined in terms of mind change bounds. The resulting model is closely related to iterative learning with bounded mind change complexity, but the dual use of mind change bounds — for hypotheses and for data — is a key distinctive feature of our approach. We show that situations exists where the more mind changes the learner is willing to accept, the lesser the amount of data it needs to remember in order to converge to the correct hypothesis. We also investigate relationships between our model and learning from good examples, set-driven, monotonic and strong-monotonic learners, as well as class-comprising versus class-preserving learnability.

Acoustic keyword spotting in speech with applications to data mining

Keyword Spotting is the task of detecting keywords of interest within continu- ous speech. The applications of this technology range from call centre dialogue systems to covert speech surveillance devices. Keyword spotting is particularly well suited to data mining tasks such as real-time keyword monitoring and unre- stricted vocabulary audio document indexing. However, to date, many keyword spotting approaches have su®ered from poor detection rates, high false alarm rates, or slow execution times, thus reducing their commercial viability. This work investigates the application of keyword spotting to data mining tasks. The thesis makes a number of major contributions to the ¯eld of keyword spotting. The ¯rst major contribution is the development of a novel keyword veri¯cation method named Cohort Word Veri¯cation. This method combines high level lin- guistic information with cohort-based veri¯cation techniques to obtain dramatic improvements in veri¯cation performance, in particular for the problematic short duration target word class. The second major contribution is the development of a novel audio document indexing technique named Dynamic Match Lattice Spotting. This technique aug- ments lattice-based audio indexing principles with dynamic sequence matching techniques to provide robustness to erroneous lattice realisations. The resulting algorithm obtains signi¯cant improvement in detection rate over lattice-based audio document indexing while still maintaining extremely fast search speeds. The third major contribution is the study of multiple veri¯er fusion for the task of keyword veri¯cation. The reported experiments demonstrate that substantial improvements in veri¯cation performance can be obtained through the fusion of multiple keyword veri¯ers. The research focuses on combinations of speech background model based veri¯ers and cohort word veri¯ers. The ¯nal major contribution is a comprehensive study of the e®ects of limited training data for keyword spotting. This study is performed with consideration as to how these e®ects impact the immediate development and deployment of speech technologies for non-English languages.