Exponentiated gradient algorithms for conditional random fields and max-margin Markov Networks

Log-linear and maximum-margin models are two commonly-used methods in supervised machine learning, and are frequently used in structured prediction problems. Efficient learning of parameters in these models is therefore an important problem, and becomes a key factor when learning from very large data sets. This paper describes exponentiated gradient (EG) algorithms for training such models, where EG updates are applied to the convex dual of either the log-linear or max-margin objective function; the dual in both the log-linear and max-margin cases corresponds to minimizing a convex function with simplex constraints. We study both batch and online variants of the algorithm, and provide rates of convergence for both cases. In the max-margin case, O(1/ε) EG updates are required to reach a given accuracy ε in the dual; in contrast, for log-linear models only O(log(1/ε)) updates are required. For both the max-margin and log-linear cases, our bounds suggest that the online EG algorithm requires a factor of n less computation to reach a desired accuracy than the batch EG algorithm, where n is the number of training examples. Our experiments confirm that the online algorithms are much faster than the batch algorithms in practice. We describe how the EG updates factor in a convenient way for structured prediction problems, allowing the algorithms to be efficiently applied to problems such as sequence learning or natural language parsing. We perform extensive evaluation of the algorithms, comparing them to L-BFGS and stochastic gradient descent for log-linear models, and to SVM-Struct for max-margin models. The algorithms are applied to a multi-class problem as well as to a more complex large-scale parsing task. In all these settings, the EG algorithms presented here outperform the other methods.

Tourism Destination Preferences: A Longitudinal Analysis of Consumer Decision Sets

There have only been a small number of applications of consumer decision set theory to holiday destination choice, and these studies have tended to rely on a single cross sectional snapshot of research participants’ stated preferences. Very little has been reported on the relationship between stated destination preferences and actual travel, or changes in decision set composition over time. The paper presents a rare longitudinal examination of destination decision sets, in the context of short break holidays by car in Queensland, Australia. Two questionnaires were administered, three months apart. The first identified destination preferences while the second examined actual travel and revisited destination preferences. In relation to the conference theme, there was very little change in consumer preferences towards the competitive set of destinations over the three month period. A key implication for the destination of interest, which, in an attempt to change market perceptions, launched a new brand campaign during the period of the project, is that a long term investment in a consistent brand message will be required to change market perceptions. The results go some way to support the proposition that the positioning of a destination into a consumer’s decision set represents a source of competitive advantage.

Exploiting multiple feature sets in data-driven impostor dataset selection for speaker verification

This study assesses the recently proposed data-driven background dataset refinement technique for speaker verification using alternate SVM feature sets to the GMM supervector features for which it was originally designed. The performance improvements brought about in each trialled SVM configuration demonstrate the versatility of background dataset refinement. This work also extends on the originally proposed technique to exploit support vector coefficients as an impostor suitability metric in the data-driven selection process. Using support vector coefficients improved the performance of the refined datasets in the evaluation of unseen data. Further, attempts are made to exploit the differences in impostor example suitability measures from varying features spaces to provide added robustness.

Max Dupain and the photography of Australian architecture

This thesis considers Max Dupain (1911-1992) and his contribution to the development of architectural photography in Australia. Through his continuous and prolific output over six decades of professional photography Dupain greatly stimulated awareness of and interest in Australian architecture. Before Dupain began specialising in the field, little consistent professional architectural photography had been practised in Australia. He and some of his close associates subsequently developed architectural photography as both a specialised branch of photography and - perhaps more significantly - as a necessary adjunct to architectural practice. In achieving these dual accomplishments, Dupain and like-minded practitioners succeeded in elevating architectural photography to the status of a discipline in its own right. They also gave Australians generally a deeper understanding of the heritage represented by the nation's built environment. At the same time, some of the photographic images he created became firmly fixed in the public imagination as historical icons within the development of a distinctive Australian tradition in the visual arts. Within his chosen field Dupain was the dominant Australian figure of his time. He was instrumental in breaking the link with Pictorialism by bringing Modernist and Documentary perspectives to Australian architectural photography. He was an innovator in the earlier decades of his professional career, however, his photographic techniques and practice did not develop beyond that. By the end of the 1980s he had largely lost touch with the technology and techniques of contemporary practice. Dupain's reputation, which has continued growing since his death in 1992, therefore arises from reasons other than his photographic images alone. It reflects his accomplishment in raising his fellow citizens' awareness of a worthwhile home-grown artistic tradition.

Rough sets based reasoning and pattern mining for a two-stage information filtering system

This paper presents a novel two-stage information filtering model which combines the merits of term-based and pattern- based approaches to effectively filter sheer volume of information. In particular, the first filtering stage is supported by a novel rough analysis model which efficiently removes a large number of irrelevant documents, thereby addressing the overload problem. The second filtering stage is empowered by a semantically rich pattern taxonomy mining model which effectively fetches incoming documents according to the specific information needs of a user, thereby addressing the mismatch problem. The experiments have been conducted to compare the proposed two-stage filtering (T-SM) model with other possible "term-based + pattern-based" or "term-based + term-based" IF models. The results based on the RCV1 corpus show that the T-SM model significantly outperforms other types of "two-stage" IF models.

The borders that law sets on entertainment

It is natural for those involved in entertainment to focus on the art. However, like any activity in even a free society, those involved in entertainment industries must operate within borders set by the law. This article examines the main areas of law that impact entertainment in an Australian context. It contrasts the position in relation to freedom of expression in Australia with that in the United States, which also promotes freedom of expression in a free society. It then briefly canvases the main limits on entertainment productions under Australian law.

Heterogeneity in auditory alarm sets makes them easier to learn

The primary objective of the experiments reported here was to demonstrate the effects of opening up the design envelope for auditory alarms on the ability of people to learn the meanings of a set of alarms. Two sets of alarms were tested, one already extant and one newly-designed set for the same set of functions, designed according to a rationale set out by the authors aimed at increasing the heterogeneity of the alarm set and incorporating some well-established principles of alarm design. For both sets of alarms, a similarity-rating experiment was followed by a learning experiment. The results showed that the newly-designed set was judged to be more internally dissimilar, and easier to learn, than the extant set. The design rationale outlined in the paper is useful for design purposes in a variety of practical domains and shows how alarm designers, even at a relatively late stage in the design process, can improve the efficacy of an alarm set.

Probabilistic matching of image sets for video-based face recognition

We address the problem of face recognition on video by employing the recently proposed probabilistic linear discrimi-nant analysis (PLDA). The PLDA has been shown to be robust against pose and expression in image-based face recognition. In this research, the method is extended and applied to video where image set to image set matching is performed. We investigate two approaches of computing similarities between image sets using the PLDA: the closest pair approach and the holistic sets approach. To better model face appearances in video, we also propose the heteroscedastic version of the PLDA which learns the within-class covariance of each individual separately. Our experi-ments on the VidTIMIT and Honda datasets show that the combination of the heteroscedastic PLDA and the closest pair approach achieves the best performance.