Modeling Efficient Classification as a Process of Confidence Assessment and Delegation

In visual object detection and recognition, classifiers have two interesting characteristics: accuracy and speed. Accuracy depends on the complexity of the image features and classifier decision surfaces. Speed depends on the hardware and the computational effort required to use the features and decision surfaces. When attempts to increase accuracy lead to increases in complexity and effort, it is necessary to ask how much are we willing to pay for increased accuracy. For example, if increased computational effort implies quickly diminishing returns in accuracy, then those designing inexpensive surveillance applications cannot aim for maximum accuracy at any cost. It becomes necessary to find trade-offs between accuracy and effort. We study efficient classification of images depicting real-world objects and scenes. Classification is efficient when a classifier can be controlled so that the desired trade-off between accuracy and effort (speed) is achieved and unnecessary computations are avoided on a per input basis. A framework is proposed for understanding and modeling efficient classification of images. Classification is modeled as a tree-like process. In designing the framework, it is important to recognize what is essential and to avoid structures that are narrow in applicability. Earlier frameworks are lacking in this regard. The overall contribution is two-fold. First, the framework is presented, subjected to experiments, and shown to be satisfactory. Second, certain unconventional approaches are experimented with. This allows the separation of the essential from the conventional. To determine if the framework is satisfactory, three categories of questions are identified: trade-off optimization, classifier tree organization, and rules for delegation and confidence modeling. Questions and problems related to each category are addressed and empirical results are presented. For example, related to trade-off optimization, we address the problem of computational bottlenecks that limit the range of trade-offs. We also ask if accuracy versus effort trade-offs can be controlled after training. For another example, regarding classifier tree organization, we first consider the task of organizing a tree in a problem-specific manner. We then ask if problem-specific organization is necessary.

A data fusion approach of multiple maintenance data sources for real-world reliability modelling

A central tenet in the theory of reliability modelling is the quantification of the probability of asset failure. In general, reliability depends on asset age and the maintenance policy applied. Usually, failure and maintenance times are the primary inputs to reliability models. However, for many organisations, different aspects of these data are often recorded in different databases (e.g. work order notifications, event logs, condition monitoring data, and process control data). These recorded data cannot be interpreted individually, since they typically do not have all the information necessary to ascertain failure and preventive maintenance times. This paper presents a methodology for the extraction of failure and preventive maintenance times using commonly-available, real-world data sources. A text-mining approach is employed to extract keywords indicative of the source of the maintenance event. Using these keywords, a Naïve Bayes classifier is then applied to attribute each machine stoppage to one of two classes: failure or preventive. The accuracy of the algorithm is assessed and the classified failure time data are then presented. The applicability of the methodology is demonstrated on a maintenance data set from an Australian electricity company.

Combat sports analytics: Boxing punch classification using overhead depth imagery

In competitive combat sporting environments like boxing, the statistics on a boxer's performance, including the amount and type of punches thrown, provide a valuable source of data and feedback which is routinely used for coaching and performance improvement purposes. This paper presents a robust framework for the automatic classification of a boxer's punches. Overhead depth imagery is employed to alleviate challenges associated with occlusions, and robust body-part tracking is developed for the noisy time-of-flight sensors. Punch recognition is addressed through both a multi-class SVM and Random Forest classifiers. A coarse-to-fine hierarchical SVM classifier is presented based on prior knowledge of boxing punches. This framework has been applied to shadow boxing image sequences taken at the Australian Institute of Sport with 8 elite boxers. Results demonstrate the effectiveness of the proposed approach, with the hierarchical SVM classifier yielding a 96% accuracy, signifying its suitability for analysing athletes punches in boxing bouts.

Robust and High-resolution Voiced/Unvoiced Classification in Noisy Speech Using A Signal Smoothness Criterion

We propose a novel technique for robust voiced/unvoiced segment detection in noisy speech, based on local polynomial regression. The local polynomial model is well-suited for voiced segments in speech. The unvoiced segments are noise-like and do not exhibit any smooth structure. This property of smoothness is used for devising a new metric called the variance ratio metric, which, after thresholding, indicates the voiced/unvoiced boundaries with 75% accuracy for 0dB global signal-to-noise ratio (SNR). A novelty of our algorithm is that it processes the signal continuously, sample-by-sample rather than frame-by-frame. Simulation results on TIMIT speech database (downsampled to 8kHz) for various SNRs are presented to illustrate the performance of the new algorithm. Results indicate that the algorithm is robust even in high noise levels.

Meltdown - a tool for classification and analysis of DSF data

An application that translates raw thermal melt curve data into more easily assimilated knowledge is described. This program, called ‘Meltdown’, performs a number of data remediation steps before classifying melt curves and estimating melting temperatures. The final output is a report that summarizes the results of a differential scanning fluorimetry experiment. Meltdown uses a Bayesian classification scheme, enabling reproducible identification of various trends commonly found in DSF datasets. The goal of Meltdown is not to replace human analysis of the raw data, but to provide a sensible interpretation of the data to make this useful experimental technique accessible to naïve users, as well as providing a starting point for detailed analyses by more experienced users.

A Reactive Inverse PN Algorithm for Collision Avoidance among Multiple Unmanned Aerial Vehicles

While performing a mission, multiple Unmanned Aerial Vehicles (UAVs) need to avoid each other to prevent collisions among them. In this paper, we design a collision avoidance algorithm to resolve the conflict among UAVs that are on a collision course while flying to heir respective destinations. The collision avoidance algorithm consist of each UAV that is on a collision course reactively executing a maneuver that will, as in `inverse' Proportional Navigation (PN), increase Line of Sight (LOS) rate between them, resulting in a `pulling out' of collision course. The algorithm is tested for high density traffic scenarios as well as for robustness in the presence of noise.

Studies on Plant Demography: Ranunculus Repens L., R. Bulbosus L. and R. Acris L.: III. A Mathematical Model Incorporating Multiple Modes of Reproduction

This paper presents a comparative population dynamics study of three closely related species of buttercups (Ranunculus repens, R. acris, and R. bulbosus). The study is based on an investigation of the behaviour of the seeds in soil under field conditions and a continuous monitoring of survival and reproduction of some 9000 individual plants over a period of 21/2 years in a coastal grassland in North Wales. The data were analysed with the help of an extension of Leslie's matrix method which makes possible an simultaneous treatment of vegetative and sexual reproduction. It was found that R. repens (a) depends more heavily on vegetative as compared with sexual reproduction, (b) shows indications of negatively density-dependent population regulation, and (c) exhibits little variation in population growth rates from site to site and from one year to the next. In contrast, R. bulbosus (a) depends exclusively on sexual reproduction, (b) shows indications of a positively density-dependent population behaviour, and (c) exhibits great variation in population growth rates from site to site and from one year to the next. R. acris exhibits an intermediate behaviour in all these respects. It is suggested that the attributes of R. repens are those expected of a species inhabiting a stable environment, while R. bulbosus exhibits some of the characteristics of a fugitive species.

Feature extraction for fingerprint classification

This paper presents two algorithms for smoothing and feature extraction for fingerprint classification. Deutsch's(2) Thinning algorithm (rectangular array) is used for thinning the digitized fingerprint (binary version). A simple algorithm is also suggested for classifying the fingerprints. Experimental results obtained using such algorithms are presented.

Studies of multiple ionization in the RF spark and vacuum vibrator arc ion sources

Relative abundance distributions of multiply-charged ionic species have been measured for the RF spark and vacuum vibrator are ion sources, for a number of elements. An attempt has been made to explain the observed charge state distribution on the basis of models for the arc and spark plasma. The difficulties in the way of explaining the observed charge state distributions, using the LTE model with Saha distribution as well as the corona model, are pointed out. The distribution can be explained by a diffusion-dominated plasma model with known or calculated values for ionization cross-sections, the single impact model being suitable for the RF spark and the multiple impact model for the vibrator arc.

Human action recognition from video sequences

This PhD research has proposed new machine learning techniques to improve human action recognition based on local features. Several novel video representation and classification techniques have been proposed to increase the performance with lower computational complexity. The major contributions are the construction of new feature representation techniques, based on advanced machine learning techniques such as multiple instance dictionary learning, Latent Dirichlet Allocation (LDA) and Sparse coding. A Binary-tree based classification technique was also proposed to deal with large amounts of action categories. These techniques are not only improving the classification accuracy with constrained computational resources but are also robust to challenging environmental conditions. These developed techniques can be easily extended to a wide range of video applications to provide near real-time performance.

Classification of Nonenzymatic Homologues of Protein Kinases

Protein Kinase-Like Non-kinases (PKLNKs), which are closely related to protein kinases, lack the crucial catalytic aspartate in the catalytic loop, and hence cannot function as protein kinase, have been analysed. Using various sensitive sequence analysis methods, we have recognized 82 PKLNKs from four higher eukaryotic organisms, namely, Homo sapiens, Mus musculus, Rattus norvegicus, and Drosophila melanogaster. On the basis of their domain combination and function, PKLNKs have been classified mainly into four categories: (1) Ligand binding PKLNKs, (2) PKLNKs with extracellular protein-protein interaction domain, (3) PKLNKs involved in dimerization, and (4) PKLNKs with cytoplasmic protein-protein interaction module. While members of the first two classes of PKLNKs have transmembrane domain tethered to the PKLNK domain, members of the other two classes of PKLNKs are cytoplasmic in nature. The current classification scheme hopes to provide a convenient framework to classify the PKLNKs from other eukaryotes which would be helpful in deciphering their roles in cellular processes.

Dancing practices from multiple perspectives: the increasingly complex role of the contemporary dancer

This paper will discuss the complexities of the role of contemporary dancer in this current epoch, with a particular focus on the multiple identities dancers embody within dance practice and how these accumulate to form a creative self-in-process or ‘moving identity’. Wider issues, such as training will be explored questioning how technical skills can be imparted alongside autonomous learning approaches to ensure that dancers are prepared to negotiate the entrepreneurial ecology of various dance sectors. Furthermore, the paper will examine the shifting relationship between choreographer and dancer from hierarchical to co-creative including how, in spite of the often collaborative nature of dance creation, the marketplace continues to celebrate the singular authorial position of the choreographer. Each of these elements will reflect back the complex issues of agency and creative self-hood that dancers must negotiate in an increasingly diverse and changeable arts environment.

Phylogeography and Adaptive Divergence of Three-spined Stickleback Populations

Genetic studies on phylogeography and adaptive divergence in Northern Hemisphere fish species such as three-spined stickleback (Gasterosteus aculeatus) provide an excellent opportunity to investigate genetic mechanisms underlying population differentiation. According to the theory, the process of population differentiation results from a complex interplay between random and deterministic processes as well historical factors. The main scope in this thesis was to study how historical factors like the Pleistocene ice ages have shaped the patterns molecular diversity in three-spined stickleback populations in Europe and how this information could be utilized in the conservation genetic context. Furthermore, identifying footprints of natural selection at the DNA level might be used in identifying genes involved in evolutionary change. Overall, the results from phylogeographic studies indicate that the three-spined stickleback has colonized the Atlantic basin relatively recently but constitutes three major evolutionary lineages in Europe. In addition, the colonization of freshwater appears to result from multiple and independent invasions by the marine conspecifics. Molecular data together with morphology suggest that the most divergent freshwater populations are located in the Balkan Peninsula and these populations deserve a special conservation genetic status without warranting further taxonomical classification. In order to investigate the adaptive divergence in Fennoscandian three-spined stickleback populations several approaches were used. First, sequence variability in the Eda-gene, coding for the number of lateral plates, was concordant with the previously observed global pattern. Full plated allele is in high frequencies among marine populations whereas low plated allele dominates in the freshwater populations. Second, a microsatellite based genome scan identified both indications of balancing and directional selection in the three-spined stickleback genome, i.e. loci with unusually similar or unusually different allele frequencies over populations. The directionally selected loci were mainly associated with the adaptation to freshwater. A follow up study conducting a more detailed analysis in a chromosome region containing a putatively selected gene locus identified a fairly large genomic region affected by natural selection. However, this region contained several gene predictions, all of which might be the actual target of natural selection. All in all, the phylogeographic and adaptive divergence studies indicate that most of the genetic divergence has occurred in the freshwater populations whereas the marine populations have remained relatively uniform.