This thesis research attempts to observe, measure and visualize the communication patterns among developers of an open source community and analyze how this can be inferred in terms of progress of that open source project. Here I attempted to analyze the Ubuntu open source project's email data (9 subproject log archives over a period of five years) and focused on drawing more precise metrics from different perspectives of the communication data. Also, I attempted to overcome the scalability issue by using Apache Pig libraries, which run on a MapReduce framework based Hadoop Cluster. I described four metrics based on which I observed and analyzed the data and also presented the results which show the required patterns and anomalies to better understand and infer the communication. Also described the usage experience with Pig Latin (scripting language of Apache Pig Libraries) for this research and how they brought the feature of scalability, simplicity, and visibility in this data intensive research work. These approaches are useful in project monitoring, to augment human observation and reporting, in social network analysis, to track individual contributions.

Real-world environments are characterized by non-stationary and continuously evolving data. Learning a classification model on this data would require a framework that is able to adapt itself to newer circumstances. Under such circumstances, transfer learning has come to be a dependable methodology for improving classification performance with reduced training costs and without the need for explicit relearning from scratch. In this thesis, a novel instance transfer technique that adapts a "Cost-sensitive" variation of AdaBoost is presented. The method capitalizes on the theoretical and functional properties of AdaBoost to selectively reuse outdated training instances obtained from a "source" domain to effectively classify unseen instances occurring in a different, but related "target" domain. The algorithm is evaluated on real-world classification problems namely accelerometer based 3D gesture recognition, smart home activity recognition and text categorization. The performance on these datasets is analyzed and evaluated against popular boosting-based instance transfer techniques. In addition, supporting empirical studies, that investigate some of the less explored bottlenecks of boosting based instance transfer methods, are presented, to understand the suitability and effectiveness of this form of knowledge transfer.

Multi-label learning, which deals with data associated with multiple labels simultaneously, is ubiquitous in real-world applications. To overcome the curse of dimensionality in multi-label learning, in this thesis I study multi-label dimensionality reduction, which extracts a small number of features by removing the irrelevant, redundant, and noisy information while considering the correlation among different labels in multi-label learning. Specifically, I propose Hypergraph Spectral Learning (HSL) to perform dimensionality reduction for multi-label data by exploiting correlations among different labels using a hypergraph. The regularization effect on the classical dimensionality reduction algorithm known as Canonical Correlation Analysis (CCA) is elucidated in this thesis. The relationship between CCA and Orthonormalized Partial Least Squares (OPLS) is also investigated. To perform dimensionality reduction efficiently for large-scale problems, two efficient implementations are proposed for a class of dimensionality reduction algorithms, including canonical correlation analysis, orthonormalized partial least squares, linear discriminant analysis, and hypergraph spectral learning. The first approach is a direct least squares approach which allows the use of different regularization penalties, but is applicable under a certain assumption; the second one is a two-stage approach which can be applied in the regularization setting without any assumption. Furthermore, an online implementation for the same class of dimensionality reduction algorithms is proposed when the data comes sequentially. A Matlab toolbox for multi-label dimensionality reduction has been developed and released. The proposed algorithms have been applied successfully in the Drosophila gene expression pattern image annotation. The experimental results on some benchmark data sets in multi-label learning also demonstrate the effectiveness and efficiency of the proposed algorithms.

Multi-task learning (MTL) aims to improve the generalization performance (of the resulting classifiers) by learning multiple related tasks simultaneously. Specifically, MTL exploits the intrinsic task relatedness, based on which the informative domain knowledge from each task can be shared across multiple tasks and thus facilitate the individual task learning. It is particularly desirable to share the domain knowledge (among the tasks) when there are a number of related tasks but only limited training data is available for each task. Modeling the relationship of multiple tasks is critical to the generalization performance of the MTL algorithms. In this dissertation, I propose a series of MTL approaches which assume that multiple tasks are intrinsically related via a shared low-dimensional feature space. The proposed MTL approaches are developed to deal with different scenarios and settings; they are respectively formulated as mathematical optimization problems of minimizing the empirical loss regularized by different structures. For all proposed MTL formulations, I develop the associated optimization algorithms to find their globally optimal solution efficiently. I also conduct theoretical analysis for certain MTL approaches by deriving the globally optimal solution recovery condition and the performance bound. To demonstrate the practical performance, I apply the proposed MTL approaches on different real-world applications: (1) Automated annotation of the Drosophila gene expression pattern images; (2) Categorization of the Yahoo web pages. Our experimental results demonstrate the efficiency and effectiveness of the proposed algorithms.

Detecting anatomical structures, such as the carina, the pulmonary trunk and the aortic arch, is an important step in designing a CAD system of detection Pulmonary Embolism. The presented CAD system gets rid of the high-level prior defined knowledge to become a system which can easily extend to detect other anatomic structures. The system is based on a machine learning algorithm --- AdaBoost and a general feature --- Haar. This study emphasizes on off-line and on-line AdaBoost learning. And in on-line AdaBoost, the thesis further deals with extremely imbalanced condition. The thesis first reviews several knowledge-based detection methods, which are relied on human being's understanding of the relationship between anatomic structures. Then the thesis introduces a classic off-line AdaBoost learning. The thesis applies different cascading scheme, namely multi-exit cascading scheme. The comparison between the two methods will be provided and discussed. Both of the off-line AdaBoost methods have problems in memory usage and time consuming. Off-line AdaBoost methods need to store all the training samples and the dataset need to be set before training. The dataset cannot be enlarged dynamically. Different training dataset requires retraining the whole process. The retraining is very time consuming and even not realistic. To deal with the shortcomings of off-line learning, the study exploited on-line AdaBoost learning approach. The thesis proposed a novel pool based on-line method with Kalman filters and histogram to better represent the distribution of the samples' weight. Analysis of the performance, the stability and the computational complexity will be provided in the thesis. Furthermore, the original on-line AdaBoost performs badly in imbalanced conditions, which occur frequently in medical image processing. In image dataset, positive samples are limited and negative samples are countless. A novel Self-Adaptive Asymmetric On-line Boosting method is presented. The method utilized a new asymmetric loss criterion with self-adaptability according to the ratio of exposed positive and negative samples and it has an advanced rule to update sample's importance weight taking account of both classification result and sample's label. Compared to traditional on-line AdaBoost Learning method, the new method can achieve far more accuracy in imbalanced conditions.

With the introduction of compressed sensing and sparse representation,many image processing and computer vision problems have been looked at in a new way. Recent trends indicate that many challenging computer vision and image processing problems are being solved using compressive sensing and sparse representation algorithms. This thesis assays some applications of compressive sensing and sparse representation with regards to image enhancement, restoration and classication. The first application deals with image Super-Resolution through compressive sensing based sparse representation. A novel framework is developed for understanding and analyzing some of the implications of compressive sensing in reconstruction and recovery of an image through raw-sampled and trained dictionaries. Properties of the projection operator and the dictionary are examined and the corresponding results presented. In the second application a novel technique for representing image classes uniquely in a high-dimensional space for image classification is presented. In this method, design and implementation strategy of the image classification system through unique affine sparse codes is presented, which leads to state of the art results. This further leads to analysis of some of the properties attributed to these unique sparse codes. In addition to obtaining these codes, a strong classier is designed and implemented to boost the results obtained. Evaluation with publicly available datasets shows that the proposed method outperforms other state of the art results in image classication. The final part of the thesis deals with image denoising with a novel approach towards obtaining high quality denoised image patches using only a single image. A new technique is proposed to obtain highly correlated image patches through sparse representation, which are then subjected to matrix completion to obtain high quality image patches. Experiments suggest that there may exist a structure within a noisy image which can be exploited for denoising through a low-rank constraint.

Reliable extraction of human pose features that are invariant to view angle and body shape changes is critical for advancing human movement analysis. In this dissertation, the multifactor analysis techniques, including the multilinear analysis and the multifactor Gaussian process methods, have been exploited to extract such invariant pose features from video data by decomposing various key contributing factors, such as pose, view angle, and body shape, in the generation of the image observations. Experimental results have shown that the resulting pose features extracted using the proposed methods exhibit excellent invariance properties to changes in view angles and body shapes. Furthermore, using the proposed invariant multifactor pose features, a suite of simple while effective algorithms have been developed to solve the movement recognition and pose estimation problems. Using these proposed algorithms, excellent human movement analysis results have been obtained, and most of them are superior to those obtained from state-of-the-art algorithms on the same testing datasets. Moreover, a number of key movement analysis challenges, including robust online gesture spotting and multi-camera gesture recognition, have also been addressed in this research. To this end, an online gesture spotting framework has been developed to automatically detect and learn non-gesture movement patterns to improve gesture localization and recognition from continuous data streams using a hidden Markov network. In addition, the optimal data fusion scheme has been investigated for multicamera gesture recognition, and the decision-level camera fusion scheme using the product rule has been found to be optimal for gesture recognition using multiple uncalibrated cameras. Furthermore, the challenge of optimal camera selection in multi-camera gesture recognition has also been tackled. A measure to quantify the complementary strength across cameras has been proposed. Experimental results obtained from a real-life gesture recognition dataset have shown that the optimal camera combinations identified according to the proposed complementary measure always lead to the best gesture recognition results.

Bridging semantic gap is one of the fundamental problems in multimedia computing and pattern recognition. The challenge of associating low-level signal with their high-level semantic interpretation is mainly due to the fact that semantics are often conveyed implicitly in a context, relying on interactions among multiple levels of concepts or low-level data entities. Also, additional domain knowledge may often be indispensable for uncovering the underlying semantics, but in most cases such domain knowledge is not readily available from the acquired media streams. Thus, making use of various types of contextual information and leveraging corresponding domain knowledge are vital for effectively associating high-level semantics with low-level signals with higher accuracies in multimedia computing problems. In this work, novel computational methods are explored and developed for incorporating contextual information/domain knowledge in different forms for multimedia computing and pattern recognition problems. Specifically, a novel Bayesian approach with statistical-sampling-based inference is proposed for incorporating a special type of domain knowledge, spatial prior for the underlying shapes; cross-modality correlations via Kernel Canonical Correlation Analysis is explored and the learnt space is then used for associating multimedia contents in different forms; model contextual information as a graph is leveraged for regulating interactions among high-level semantic concepts (e.g., category labels), low-level input signal (e.g., spatial/temporal structure). Four real-world applications, including visual-to-tactile face conversion, photo tag recommendation, wild web video classification and unconstrained consumer video summarization, are selected to demonstrate the effectiveness of the approaches. These applications range from classic research challenges to emerging tasks in multimedia computing. Results from experiments on large-scale real-world data with comparisons to other state-of-the-art methods and subjective evaluations with end users confirmed that the developed approaches exhibit salient advantages, suggesting that they are promising for leveraging contextual information/domain knowledge for a wide range of multimedia computing and pattern recognition problems.

Genes have widely different pertinences to the etiology and pathology of diseases. Thus, they can be ranked according to their disease-significance on a genomic scale, which is the subject of gene prioritization. Given a set of genes known to be related to a disease, it is reasonable to use them as a basis to determine the significance of other candidate genes, which will then be ranked based on the association they exhibit with respect to the given set of known genes. Experimental and computational data of various kinds have different reliability and relevance to a disease under study. This work presents a gene prioritization method based on integrated biological networks that incorporates and models the various levels of relevance and reliability of diverse sources. The method is shown to achieve significantly higher performance as compared to two well-known gene prioritization algorithms. Essentially, no bias in the performance was seen as it was applied to diseases of diverse ethnology, e.g., monogenic, polygenic and cancer. The method was highly stable and robust against significant levels of noise in the data. Biological networks are often sparse, which can impede the operation of associationbased gene prioritization algorithms such as the one presented here from a computational perspective. As a potential approach to overcome this limitation, we explore the value that transcription factor binding sites can have in elucidating suitable targets. Transcription factors are needed for the expression of most genes, especially in higher organisms and hence genes can be associated via their genetic regulatory properties. While each transcription factor recognizes specific DNA sequence patterns, such patterns are mostly unknown for many transcription factors. Even those that are known are inconsistently reported in the literature, implying a potentially high level of inaccuracy. We developed computational methods for prediction and improvement of transcription factor binding patterns. Tests performed on the improvement method by employing synthetic patterns under various conditions showed that the method is very robust and the patterns produced invariably converge to nearly identical series of patterns. Preliminary tests were conducted to incorporate knowledge from transcription factor binding sites into our networkbased model for prioritization, with encouraging results. Genes have widely different pertinences to the etiology and pathology of diseases. Thus, they can be ranked according to their disease-significance on a genomic scale, which is the subject of gene prioritization. Given a set of genes known to be related to a disease, it is reasonable to use them as a basis to determine the significance of other candidate genes, which will then be ranked based on the association they exhibit with respect to the given set of known genes. Experimental and computational data of various kinds have different reliability and relevance to a disease under study. This work presents a gene prioritization method based on integrated biological networks that incorporates and models the various levels of relevance and reliability of diverse sources. The method is shown to achieve significantly higher performance as compared to two well-known gene prioritization algorithms. Essentially, no bias in the performance was seen as it was applied to diseases of diverse ethnology, e.g., monogenic, polygenic and cancer. The method was highly stable and robust against significant levels of noise in the data. Biological networks are often sparse, which can impede the operation of associationbased gene prioritization algorithms such as the one presented here from a computational perspective. As a potential approach to overcome this limitation, we explore the value that transcription factor binding sites can have in elucidating suitable targets. Transcription factors are needed for the expression of most genes, especially in higher organisms and hence genes can be associated via their genetic regulatory properties. While each transcription factor recognizes specific DNA sequence patterns, such patterns are mostly unknown for many transcription factors. Even those that are known are inconsistently reported in the literature, implying a potentially high level of inaccuracy. We developed computational methods for prediction and improvement of transcription factor binding patterns. Tests performed on the improvement method by employing synthetic patterns under various conditions showed that the method is very robust and the patterns produced invariably converge to nearly identical series of patterns. Preliminary tests were conducted to incorporate knowledge from transcription factor binding sites into our networkbased model for prioritization, with encouraging results. To validate these approaches in a disease-specific context, we built a schizophreniaspecific network based on the inferred associations and performed a comprehensive prioritization of human genes with respect to the disease. These results are expected to be validated empirically, but computational validation using known targets are very positive.

Cloud computing has received significant attention recently as it is a new computing infrastructure to enable rapid delivery of computing resources as a utility in a dynamic, scalable, and visualized manner. SaaS (Software-as-a-Service) provide a now paradigm in cloud computing, which goal is to provide an effective and intelligent way to support end users' on-demand requirements to computing resources, including maturity levels of customizable, multi-tenancy and scalability. To meet requirements of on-demand, my thesis discusses several critical research problems and proposed solutions using real application scenarios. Service providers receive multiple requests from customers, how to prioritize those service requests to maximize the business values is one of the most important issues in cloud. An innovative prioritization model is proposed, which uses different types of information, including customer, service, environment and workflow information to optimize the performance of the system. To provide "on-demand" services, an accurate demand prediction and provision become critical for the successful of the cloud computing. An effective demand prediction model is proposed, and applied to a real mortgage application. To support SaaS customization and fulfill the various functional and quality requirements of individual tenants, a unified and innovative multi-layered customization framework is proposed to support and manage the variability of SaaS applications. To support scalable SaaS, a hybrid database design to support SaaS customization with two-layer database partitioning is proposed. To support secure SaaS, O-RBAC, an ontology based RBAC (Role based Access Control) model is used for Multi-Tenancy Architecture in clouds. To support a significant number of tenants, an easy to use SaaS construction framework is proposed. As a summary, this thesis discusses the most important research problems in cloud computing, towards effective and intelligent SaaS. The research in this thesis is critical to the development of cloud computing and provides fundamental solutions to those problems.