Model selection and error estimation

We study model selection strategies based on penalized empirical loss minimization. We point out a tight relationship between error estimation and data-based complexity penalization: any good error estimate may be converted into a data-based penalty function and the performance of the estimate is governed by the quality of the error estimate. We consider several penalty functions, involving error estimates on independent test data, empirical VC dimension, empirical VC entropy, and margin-based quantities. We also consider the maximal difference between the error on the first half of the training data and the second half, and the expected maximal discrepancy, a closely related capacity estimate that can be calculated by Monte Carlo integration. Maximal discrepancy penalty functions are appealing for pattern classification problems, since their computation is equivalent to empirical risk minimization over the training data with some labels flipped.

Learning the kernel matrix with semidefinite programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive semidefinite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space - classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semidefinite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -using the labeled part of the data one can learn an embedding also for the unlabeled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method for learning the 2-norm soft margin parameter in support vector machines, solving an important open problem.

Learning the Kernel Matrix with Semi-Definite Programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive definite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space -- classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semi-definite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -- using the labelled part of the data one can learn an embedding also for the unlabelled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method to learn the 2-norm soft margin parameter in support vector machines, solving another important open problem. Finally, the novel approach presented in the paper is supported by positive empirical results.

The effect of ocular surface conditions on blink rate and completeness

A healthy human would be expected to show periodic blinks, making a brief closure of the eyelids. Most blinks are spontaneous, occurring regularly with no external stimulus. However a reflex blink can occur in response to external stimuli such as a bright light, a sudden loud noise, or an object approaching toward the eyes. A voluntary or forced blink is another type of blink in which the person deliberately closes the eyes and the lower eyelid raises to meet the upper eyelid. A complete blink, in which the upper eyelid touches the lower eyelid, contributes to the health of ocular surface by providing a fresh layer of tears as well as maintaining optical integrity by providing a smooth tear film over the cornea. The rate of blinking and its completeness vary depending on the task undertaken during blink assessment, the direction of gaze, the emotional state of the subjects and the method under which the blink was measured. It is also well known that wearing contact lenses (both rigid and soft lenses) can induce significant changes in blink rate and completeness. It is been established that efficient blinking plays an important role in ocular surface health during contact lens wear and for improving contact lens performance and comfort. Inefficient blinking during contact lens wear may be related to a low blink rate or incomplete blinking and can often be a reason for dry eye symptoms or ocular surface staining. It has previously been shown that upward gaze can affect blink rate, causing it to become faster. In the first experiment, it was decided to expand on previous studies in this area by examining the effect of various gaze directions (i.e. upward gaze, primary gaze, downward gaze and lateral gaze) as well as head angle (recumbent position) on normal subjects’ blink rate and completeness through the use of filming with a high-speed camera. The results of this experiment showed that as the open palpebral aperture (and exposed ocular surface area) increased from downward gaze to upward gaze, the number of blinks significantly increased (p<0.04). Also, the size of closed palpebral aperture significantly increased from downward gaze to upward gaze (p<0.005). A weak positive correlation (R² = 0.18) between the blink rate and ocular surface area was found in this study. Also, it was found that the subjects showed 81% complete blinks, 19% incomplete blinks and 2% of twitch blinks in primary gaze, consistent with previous studies. The difference in the percentage of incomplete blinks between upward gaze and downward gaze was significant (p<0.004), showing more incomplete blinks in upward gaze. The findings of this experiment suggest that while blink rate becomes slower in downward gaze, the completeness of blinking is typically better, thereby potentially reducing the risk of tear instability. On the other hand, in upward gaze while the completeness of blinking becomes worse, this is potentially offset by increased blink frequency. In addition, blink rate and completeness were not affected by lateral gaze or head angle, possibly because these conditions have similar size of the open palpebral aperture compared with primary gaze. In the second experiment, an investigation into the changes in blink rate and completeness was carried out in primary gaze and downward gaze with soft and rigid contact lenses in unadapted wearers. Not surprisingly, rigid lens wear caused a significant increase in the blink rate in both primary (p<0.001) and downward gaze (p<0.02). After fitting rigid contact lenses, the closed palpebral aperture (blink completeness) did not show any changes but the open palpebral aperture showed a significant narrowing (p<0.04). This might occur from the subjects’ attempt to avoid interaction between the upper eyelid and the edge of the lens to minimize discomfort. After applying topical anaesthetic eye drops in the eye fitted with rigid lenses, the increased blink rate dropped to values similar to that before lens insertion and the open palpebral aperture returned to baseline values, suggesting that corneal and/or lid margin sensitivity was mediating the increased blink rate and narrowed palpebral aperture. We also investigated the changes in the blink rate and completeness with soft contact lenses including a soft sphere, double slab-off toric design and periballast toric design. Soft contact lenses did not cause any significant changes in the blink rate, closed palpebral aperture, open palpebral aperture and the percentage of incomplete blinks in either primary gaze or downward gaze. After applying anaesthetic eye drops, the blink rate reduced in both primary gaze and downward gaze, however this difference was not statistically significant. The size of the closed palpebral aperture and open palpebral aperture did not show any significant changes after applying anaesthetic eye drops. However it should be noted that the effects of rigid and soft contact lenses that we observed in these studies were only the immediate reaction to contact lenses and in the longer term, it is likely that these responses will vary as the eye adapts to the presence of the lenses.

Adaptive cache aware multiprocessor scheduling framework

Computer resource allocation represents a significant challenge particularly for multiprocessor systems, which consist of shared computing resources to be allocated among co-runner processes and threads. While an efficient resource allocation would result in a highly efficient and stable overall multiprocessor system and individual thread performance, ineffective poor resource allocation causes significant performance bottlenecks even for the system with high computing resources. This thesis proposes a cache aware adaptive closed loop scheduling framework as an efficient resource allocation strategy for the highly dynamic resource management problem, which requires instant estimation of highly uncertain and unpredictable resource patterns. Many different approaches to this highly dynamic resource allocation problem have been developed but neither the dynamic nature nor the time-varying and uncertain characteristics of the resource allocation problem is well considered. These approaches facilitate either static and dynamic optimization methods or advanced scheduling algorithms such as the Proportional Fair (PFair) scheduling algorithm. Some of these approaches, which consider the dynamic nature of multiprocessor systems, apply only a basic closed loop system; hence, they fail to take the time-varying and uncertainty of the system into account. Therefore, further research into the multiprocessor resource allocation is required. Our closed loop cache aware adaptive scheduling framework takes the resource availability and the resource usage patterns into account by measuring time-varying factors such as cache miss counts, stalls and instruction counts. More specifically, the cache usage pattern of the thread is identified using QR recursive least square algorithm (RLS) and cache miss count time series statistics. For the identified cache resource dynamics, our closed loop cache aware adaptive scheduling framework enforces instruction fairness for the threads. Fairness in the context of our research project is defined as a resource allocation equity, which reduces corunner thread dependence in a shared resource environment. In this way, instruction count degradation due to shared cache resource conflicts is overcome. In this respect, our closed loop cache aware adaptive scheduling framework contributes to the research field in two major and three minor aspects. The two major contributions lead to the cache aware scheduling system. The first major contribution is the development of the execution fairness algorithm, which degrades the co-runner cache impact on the thread performance. The second contribution is the development of relevant mathematical models, such as thread execution pattern and cache access pattern models, which in fact formulate the execution fairness algorithm in terms of mathematical quantities. Following the development of the cache aware scheduling system, our adaptive self-tuning control framework is constructed to add an adaptive closed loop aspect to the cache aware scheduling system. This control framework in fact consists of two main components: the parameter estimator, and the controller design module. The first minor contribution is the development of the parameter estimators; the QR Recursive Least Square(RLS) algorithm is applied into our closed loop cache aware adaptive scheduling framework to estimate highly uncertain and time-varying cache resource patterns of threads. The second minor contribution is the designing of a controller design module; the algebraic controller design algorithm, Pole Placement, is utilized to design the relevant controller, which is able to provide desired timevarying control action. The adaptive self-tuning control framework and cache aware scheduling system in fact constitute our final framework, closed loop cache aware adaptive scheduling framework. The third minor contribution is to validate this cache aware adaptive closed loop scheduling framework efficiency in overwhelming the co-runner cache dependency. The timeseries statistical counters are developed for M-Sim Multi-Core Simulator; and the theoretical findings and mathematical formulations are applied as MATLAB m-file software codes. In this way, the overall framework is tested and experiment outcomes are analyzed. According to our experiment outcomes, it is concluded that our closed loop cache aware adaptive scheduling framework successfully drives co-runner cache dependent thread instruction count to co-runner independent instruction count with an error margin up to 25% in case cache is highly utilized. In addition, thread cache access pattern is also estimated with 75% accuracy.

Neoproterozoic glacial deposits of the Kimberly Region and northwestern Northern Territory, Australia

Neoproterozoic glacigenic formations are preserved in the Kimberley region and northwestern Northern Territory of northern Australia. They are distributed in the west Kimberley adjacent to the northern margins of the King Leopold Orogen, the Mt Ramsay area at the junction of the King Leopold and Halls Creek Orogens, and the east Kimberley, adjacent to the eastern margin of the Halls Creek Orogen. Small outlier glacigenic deposits are preserved in the Litchfield Province, Northern Territory (Uniya Formation) and Georgina Basin, western Queensland (Little Burke Formation). Glacigenic strata comprise diamictite, conglomerate, sandstone and pebbly mudstone and characterize the Walsh, Landrigan and Fargoo/Moonlight Valley formations. Thin units of laminated dolomite sit conformably at the top of the Walsh, Landrigan and Moonlight Valley formations. Glacigenic units are also interbedded with the carbonate platform deposits of the Egan Formation and Boonall Dolomite. δ13C data are available for all carbonate units. There is no direct chronological constraint on these successions. Dispute over regional correlation of the Neoproterozoic succession has been largely resolved through biostratigraphic, chemostratigraphic and lithostratigraphic analysis. However, palaeomagnetic results from the Walsh Formation are inconsistent with sedimentologically based correlations. Two stratigraphically defined glaciations are preserved in northwestern Australia: the ‘Landrigan Glaciation’, characterized by southwest-directed continental ice-sheet movement and correlated with late Cryogenian glaciation elsewhere in Australia and the world; and, the ‘Egan Glaciation’, a more localized glaciation of the Ediacaran Period. Future research focus should include chronology, palaeomagnetic constraint and tectonostratigraphic controls on deposition.

Improved facial expression recognition via uni-hyperplane classification

Large margin learning approaches, such as support vector machines (SVM), have been successfully applied to numerous classification tasks, especially for automatic facial expression recognition. The risk of such approaches however, is their sensitivity to large margin losses due to the influence from noisy training examples and outliers which is a common problem in the area of affective computing (i.e., manual coding at the frame level is tedious so coarse labels are normally assigned). In this paper, we leverage the relaxation of the parallel-hyperplanes constraint and propose the use of modified correlation filters (MCF). The MCF is similar in spirit to SVMs and correlation filters, but with the key difference of optimizing only a single hyperplane. We demonstrate the superiority of MCF over current techniques on a battery of experiments.

Risk control in transmission system expansion planning with wind generators

With the advent of large-scale wind farms and their integration into electrical grids, more uncertainties, constraints and objectives must be considered in power system development. It is therefore necessary to introduce risk-control strategies into the planning of transmission systems connected with wind power generators. This paper presents a probability-based multi-objective model equipped with three risk-control strategies. The model is developed to evaluate and enhance the ability of the transmission system to protect against overload risks when wind power is integrated into the power system. The model involves: (i) defining the uncertainties associated with wind power generators with probability measures and calculating the probabilistic power flow with the combined use of cumulants and Gram-Charlier series; (ii) developing three risk-control strategies by specifying the smallest acceptable non-overload probability for each branch and the whole system, and specifying the non-overload margin for all branches in the whole system; (iii) formulating an overload risk index based on the non-overload probability and the non-overload margin defined; and (iv) developing a multi-objective transmission system expansion planning (TSEP) model with the objective functions composed of transmission investment and the overload risk index. The presented work represents a superior risk-control model for TSEP in terms of security, reliability and economy. The transmission expansion planning model with the three risk-control strategies demonstrates its feasibility in the case study using two typical power systems

Coming to belong : a narrative analysis of international students’ experiences in an Australian university

This thesis reports on an interview study with 17 international students about their experiences of coming to belong in an Australian university. All used English as an additional language (EAL). The students’ narratives of ‘coming to belong’ are conceptualised through the theory of Bourdieu, in particular the concepts of field, capital, habitus and legitimation; and the methodological premises of critical realism’s layered ontology. The literature review argues that access to and accrual of a range of capital is critical to successful adaptation to a new educational system. This, and processes of legitimation by others in the fields, affects the senses of belonging for students of various linguistic backgrounds, of different countries of origin, studying from primary to higher education in diverse parts of the world. Data were collected by semi-structured interviews and email dialogues at three points during the students’ first year of study in Australia. The analysis shows how the students’ empirical experiences were ordered in terms of narrative structure—orientation, complication, evaluation, resolution and coda—and highlight the emotions generated by the sequence of events. The findings show that EAL international students sought new field positions through legitimation in multiple senses across (sub-)fields. They also show that academic, social and linguistic legitimacy granted by others produced a spectrum of belonging: in the centre, at the margin, and/or to meaningful intercultural encounters. This study makes a contribution to the growing literature around the experience of international students in higher education, and to empirical literature using Bourdieu to understand educational relations.

The verdine facies off French Guiana

This chapter reviews green grains from the shelf of French Guiana as a regional example of sedimentologic process occurring on the whole stable continental margin from the Amazon to the Orinoco River. Green grains have been observed and analyzed off the Orinoco delta and on the continental shelf of Surinam. These green grains were identified as “chamosite” and “glauconite.” The muddy coast of French Guiana is generally very flat and occupied by wet swamps and mangrove as a result of the equatorial climate. Most green grains on the continental shelf represent the verdine facies. Green grains are ubiquitous on the shelf and top of the slope off French Guiana. Two sedimentological facies exist: glaucony deeper than 150 m and verdine at shallower depths. The verdine facies has mainly developed from mineral debris and especially chloritized biotite. Carbonate bioclasts and faecal pellets are also utilized. The mica flakes were never wholly replaced by authigenic clay and the phenomenon leads to mixed grains where authigenic and substrate remains are recognizable. Carbonate substrates lead to mainly clay pure green grains becasue the initial carbonate has been dissolved. The formation of verdine can be located in a general marine environment at a comparatively warm sea-water temperature and at a depth probably shallower than 60 m.

Seriously playful : the uses of networked spoof videos in China

This thesis examines online spoof videos in China. It shows the relationship between user-created content and change and how such videos are impacting on social memory. In the West, we are witnessing two outstanding trends in media. On the one hand, media are turning more "demotic" (Turner, 2006) and "participatory" (Jenkins, 2006), whereby lay audiences use popular media for identity formation, representation and association, reconfiguring the media and cultural landscape, and rendering invalid the old paradigm based on the dichotomy of audience and author, creator and consumer, expert and amateur. On the other hand, in both mainstream media and user-creation online there is a trend towards "silly citizenship", with comedy, send-ups and spoofs that used to reside in the margin propelled to the central stage in both pleasure and politics (Hartley, 2010), as is shown in the rising popularity of the Daily Show, Colbert Report, and spoof videos in elections ,e.g. the 2008 presidential election in US (Gray, Jones, & Thompson, 2009; Tryon, 2008). User generated content—and spoof subcultures—is now much a debated phenomenon in China. However, with different political (one party rule and censorship) and cultural (media regarded mainly as instrument for education and social stabilization instead of a critical fourth estate) configurations, will the social and cultural impacts of the two trends in the West be as the same in China? If not, what will be the specificities in the China context? The project starts with a historical review of popular culture and user-created content in China, before turning to spoof videos and looking at how they are produced and shared, travel and diffused on the Internet, and how the communities and sub-cultures forming and emerging around spoof videos are changing the overall cultural landscape in China. By acting as a participant observer in online video sharing sites and conducting face-to face as well as online interviews, I identify lead users and creators of spoof videos and the social networks emerging around them. I call these lead users "skill hubs" and their networks "liquid communities", foregrounding the fact that their appeal doesn’t come from their amicable personality, but rather from their creative skills; and that the networks surrounding them are in a permanent flux, with members coming and going as they see fit. I argue that the "liquidness" (Bauman, 2000) of these communities is what makes them constantly creative and appealing. Textual analysis of online videos, their comments and derivatives are conducted to tease out the uses that that can be made of spoof videos, namely as phatic communication, as alternative memory and as political engagement. Through these analyses I show that spoof videos constitute not only a space where young generations can engage with each other, communicate their anger and dissatisfaction, fun and hope, and where they participate in socio-cultural and political debates, but also create a space where they can experiment with their new skills, new ideas, and new citizenship. The rise of spoof videos heralds the beginning of a trend in popular culture in contemporary China towards the "canonization of the jester" and the dethroning of the establishment. I also argue that a historical perspective is needed to understand the current surge of use creativity and user activism in China, and that many forms of popular media we experience today have their antecedents in various stages of Chinese history. The entrenched "control-resistance" binary is inadequate in interpreting the rich, flux and multilayered Internet space in China.

Assessment of recycled concrete aggregate for road base and sub-base

Population increase and economic developments can lead to construction as well as demolition of infrastructures such as buildings, bridges, roads, etc and used concrete is the main waste product of them. Recycling of waste concrete to obtain the recycled concrete aggregates (RCA) for base and/or sub-base materials in road construction is a foremost application to be promoted to gain economical and sustainable benefits. As the mortar, bricks, glass and asphalt present in different constituents in RCA, it exhibits inconsistent properties and performance. In this study, six different types of RCA samples were subjected classification tests such as particle size distribution, plasticity, compaction test and California Bearing Ratio (CBR). Results were compared with those of the standard road materials used in Queensland, Australia and found that ‘RM1-100/RM3-0’ and ‘RM1-80/RM3-20’ samples are sitting in the margin of the minimum required specifications of base materials while others are lower than that.

Early-mid Cretaceous tectonic evolution of eastern Gondwana : from silicic LIP magmatism to continental rupture

The Early–mid Cretaceous marks the confluence of three major continental-scale events in eastern Gondwana: (1) the emplacement of a Silicic Large Igneous Province (LIP) near the continental margin; (2) the volcaniclastic fill, transgression and regression of a major epicontinental seaway developed over at least a quarter of the Australian continent; and (3) epeirogenic uplift, exhumation and continental rupturing culminating in the opening of the Tasman Basin c. 84 Ma. The Whitsunday Silicic LIP event had widespread impact, producing both substantial extrusive volumes of dominantly silicic pyroclastic material and coeval first-cycle volcanogenic sediment that accumulated within many eastern Australian sedimentary basins, and principally in the Great Australian Basin system (>2 Mkm3 combined volume). The final pulse of volcanism and volcanogenic sedimentation at c. 105–95 Ma coincided with epicontinental seaway regression, which shows a lack of correspondence with the global sea-level curve, and alternatively records a wider, continental-scale effect of volcanism and rift tectonism. Widespread igneous underplating related to this LIP event is evident from high paleogeothermal gradients and regional hydrothermal fluid flow detectable in the shallow crust and over a broad region. Enhanced CO2 fluxing through sedimentary basins also records indirectly, large-scale, LIP-related mafic underplating. A discrete episode of rapid crustal cooling and exhumation began c. 100–90 Ma along the length of the eastern Australian margin, related to an enhanced phase of continental rifting that was largely amagmatic, and probably a switch from wide–more narrow rift modes. Along-margin variations in detachment fault architecture produced narrow (SE Australia) and wide continental margins with marginal, submerged continental plateaux (NE Australia). Long-lived NE-trending cross-orogen lineaments controlled the switch from narrow to wide continental margin geometries.