Master of Science / Department of Psychology / Lester C. Loschky / The effect of blur/clarity contrast on selective attention was investigated in terms of how unique blur and/or clarity guides attention. Visual blur has previously been suggested to be processed preattentively using a dual-task paradigm (Loschky et al., 2014). Experiments 1 and 2 used rotated L and T visual search tasks with blur/clarity contrast being manipulated such that it was non-predictive of the target’s location. Each experiment was preceded by a legibility control study such that blurred and clear letters had similar accuracy and reaction times. This allowed for the results to be interpreted as changes in attention rather than difficulty identifying the letters because they were blurry. Results suggest that when non-predictive of target location, unique blur plays a passive role in selective attention in which it is ignored, neither capturing nor repelling attention to its spatial location, whereas unique clarity captures attention. The findings provide insight to the role that blur/clarity contrast plays in guiding visual attention, which can be implemented in visual software to help guide selective attention to critical regions of interest displayed on a computer screen.

This paper uses the Dutch panel data from LISS, Longitudinal Internet Studies for the Social Science in trying to establish if a relationship between individuals’ risk propensity and job search activity is present. When looking at employed and unemployed job seekers jointly, a positive significant relationship is present. Looking at these groups separately shows that the relationship is driven by employed job seekers. No relationship for unemployed job seekers can be identified when being looked at separate. However, when taking into account possible biases from changes in risk propensity over time as well as the classification of actively searching individuals, no relationship is present at all. We hence are cautious towards the significant estimates received that potentially could suffer from biases. Further studies should be made with a bigger sample of individuals and a continuously updated measure of risk propensity to minimizing potential bias.

by Wong, Wai-Chiu. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2000. / Includes bibliographical references (leaves 89-94). / Abstracts in English and Chinese. / Abstract --- p.ii / Acknowledgments --- p.iv / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Document Clustering --- p.2 / Chapter 1.2 --- DC-tree --- p.4 / Chapter 1.3 --- Feature Extraction --- p.5 / Chapter 1.4 --- Outline of the Thesis --- p.5 / Chapter 2 --- Related Work --- p.8 / Chapter 2.1 --- Clustering Algorithms --- p.8 / Chapter 2.1.1 --- Partitional Clustering Algorithms --- p.8 / Chapter 2.1.2 --- Hierarchical Clustering Algorithms --- p.10 / Chapter 2.2 --- Document Classification by Examples --- p.11 / Chapter 2.2.1 --- k-NN algorithm - Expert Network (ExpNet) --- p.11 / Chapter 2.2.2 --- Learning Linear Text Classifier --- p.12 / Chapter 2.2.3 --- Generalized Instance Set (GIS) algorithm --- p.12 / Chapter 2.3 --- Document Clustering --- p.13 / Chapter 2.3.1 --- B+-tree-based Document Clustering --- p.13 / Chapter 2.3.2 --- Suffix Tree Clustering --- p.14 / Chapter 2.3.3 --- Association Rule Hypergraph Partitioning Algorithm --- p.15 / Chapter 2.3.4 --- Principal Component Divisive Partitioning --- p.17 / Chapter 2.4 --- Projections for Efficient Document Clustering --- p.18 / Chapter 3 --- Background --- p.21 / Chapter 3.1 --- Document Preprocessing --- p.21 / Chapter 3.1.1 --- Elimination of Stopwords --- p.22 / Chapter 3.1.2 --- Stemming Technique --- p.22 / Chapter 3.2 --- Problem Modeling --- p.23 / Chapter 3.2.1 --- Basic Concepts --- p.23 / Chapter 3.2.2 --- Vector Model --- p.24 / Chapter 3.3 --- Feature Selection Scheme --- p.25 / Chapter 3.4 --- Similarity Model --- p.27 / Chapter 3.5 --- Evaluation Techniques --- p.29 / Chapter 4 --- Feature Extraction and Weighting --- p.31 / Chapter 4.1 --- Statistical Analysis of the Words in the Web Domain --- p.31 / Chapter 4.2 --- Zipf's Law --- p.33 / Chapter 4.3 --- Traditional Methods --- p.36 / Chapter 4.4 --- The Proposed Method --- p.38 / Chapter 4.5 --- Experimental Results --- p.40 / Chapter 4.5.1 --- Synthetic Data Generation --- p.40 / Chapter 4.5.2 --- Real Data Source --- p.41 / Chapter 4.5.3 --- Coverage --- p.41 / Chapter 4.5.4 --- Clustering Quality --- p.43 / Chapter 4.5.5 --- Binary Weight vs Numerical Weight --- p.45 / Chapter 5 --- Web Document Clustering Using DC-tree --- p.48 / Chapter 5.1 --- Document Representation --- p.48 / Chapter 5.2 --- Document Cluster (DC) --- p.49 / Chapter 5.3 --- DC-tree --- p.52 / Chapter 5.3.1 --- Tree Definition --- p.52 / Chapter 5.3.2 --- Insertion --- p.54 / Chapter 5.3.3 --- Node Splitting --- p.55 / Chapter 5.3.4 --- Deletion and Node Merging --- p.56 / Chapter 5.4 --- The Overall Strategy --- p.57 / Chapter 5.4.1 --- Preprocessing --- p.57 / Chapter 5.4.2 --- Building DC-tree --- p.59 / Chapter 5.4.3 --- Identifying the Interesting Clusters --- p.60 / Chapter 5.5 --- Experimental Results --- p.61 / Chapter 5.5.1 --- Alternative Similarity Measurement : Synthetic Data --- p.61 / Chapter 5.5.2 --- DC-tree Characteristics : Synthetic Data --- p.63 / Chapter 5.5.3 --- Compare DC-tree and B+-tree: Synthetic Data --- p.64 / Chapter 5.5.4 --- Compare DC-tree and B+-tree: Real Data --- p.66 / Chapter 5.5.5 --- Varying the Number of Features : Synthetic Data --- p.67 / Chapter 5.5.6 --- Non-Correlated Topic Web Page Collection: Real Data --- p.69 / Chapter 5.5.7 --- Correlated Topic Web Page Collection: Real Data --- p.71 / Chapter 5.5.8 --- Incremental updates on Real Data Set --- p.72 / Chapter 5.5.9 --- Comparison with the other clustering algorithms --- p.73 / Chapter 6 --- Conclusion --- p.75 / Appendix --- p.77 / Chapter A --- Stopword List --- p.77 / Chapter B --- Porter's Stemming Algorithm --- p.81 / Chapter C --- Insertion Algorithm --- p.83 / Chapter D --- Node Splitting Algorithm --- p.85 / Chapter E --- Features Extracted in Experiment 4.53 --- p.87 / Bibliography --- p.88

This thesis provides a model-theoretic semantic analysis of aspects of the LF logical framework

This thesis addresses the need for a scalable distributed solution for k-nearest-neighbor (kNN) search, a fundamental data mining task. This unsupervised method poses particular challenges on shared-nothing distributed architectures, where global information about the dataset is not available to individual machines. The distance to search for neighbors is not known a priori, and therefore a dynamic data partitioning strategy is required to guarantee that exact kNN can be found autonomously on each machine. Pivot-based partitioning has been shown to facilitate bounding of partitions, however state-of-the-art methods suffer from prohibitive data duplication (upwards of 20x the size of the dataset). In this work an innovative method for solving exact distributed kNN search called PkNN is presented. The key idea is to perform computation over several rounds, leveraging pivot-based data partitioning at each stage. Aggressive data-driven bounds limit communication costs, and a number of optimizations are designed for efficient computation. Experimental study on large real-world data (over 1 billion points) compares PkNN to the state-of-the-art distributed solution, demonstrating that the benefits of additional stages of computation in the PkNN method heavily outweigh the added I/O overhead. PkNN achieves a data duplication rate close to 1, significant speedup over previous solutions, and scales effectively in data cardinality and dimension. PkNN can facilitate distributed solutions to other unsupervised learning methods which rely on kNN search as a critical building block. As one example, a distributed framework for the Local Outlier Factor (LOF) algorithm is given. Testing on large real-world and synthetic data with varying characteristics measures the scalability of PkNN and the distributed LOF framework in data size and dimensionality.

In the health-informatics and bio-medical domains, clinicians produce an enormous amount of data which can be complex and high in dimensionality. This scenario includes visual field data, which are used for managing the second leading cause of blindness in the world: glaucoma. Visual field data are the most common type of data collected to diagnose glaucoma in patients, and usually the data consist of 54 or 76 variables (which are referred to as visual field locations). Due to the large number of variables, the six nerve fiber bundles (6NFB), which is a collection of visual field locations in groups, are the standard clusters used in visual field data to represent the physiological traits of the retina. However, with regard to classification accuracy of the data, this research proposes a technique to find other significant spatial clusters of visual field with higher classification accuracy than the 6NFB. This thesis presents a novel clustering technique, namely, Simultaneous Modelling and Clustering (SMC). SMC performs clustering on data based on classification accuracy using heuristic search techniques. The method searches a collection of significant clusters of visual field locations that indicate visual field loss progression. The aim of this research is two-fold. Firstly, SMC algorithms are developed and tested on data to investigate the effectiveness and efficiency of the method using optimisation and classification methods. Secondly, a significant clustering arrangement of visual field, which highly interrelated visual field locations to represent progression of visual field loss with high classification accuracy, is searched to complement the 6NFB in diagnosis of glaucoma. A new clustering arrangement of visual field locations can be used by medical practitioners together with the 6NFB to complement each other in diagnosis of glaucoma in patients. This research conducts extensive experiment work on both visual field and simulated data to evaluate the proposed method. The results obtained suggest the proposed method appears to be an effective and efficient method in clustering visual field data and 3 improving classification accuracy. The key contributions of this work are the novel model-based clustering of visual field data, effective and efficient algorithms for SMC, practical knowledge of visual field data in the diagnosis of glaucoma and the presentation a generic framework for modelling and clustering which is highly applicable to many other dataset/model combinations.

Bryan Chi-ho Lam. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2003. / Includes bibliographical references (leaves 163-166). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Background --- p.4 / Chapter 2.1 --- Constraint Satisfaction Problems --- p.4 / Chapter 2.2 --- Systematic Search --- p.5 / Chapter 2.3 --- Stochastic Search --- p.6 / Chapter 2.3.1 --- Overview --- p.6 / Chapter 2.3.2 --- GENET --- p.8 / Chapter 2.3.3 --- CSVC --- p.10 / Chapter 2.3.4 --- Adaptive Search --- p.12 / Chapter 2.4 --- Hybrid Approach --- p.13 / Chapter 3 --- Progressive Stochastic Search --- p.14 / Chapter 3.1 --- Progressive Stochastic Search --- p.14 / Chapter 3.1.1 --- Network Architecture --- p.15 / Chapter 3.1.2 --- Convergence Procedure --- p.16 / Chapter 3.1.3 --- An Illustrative Example --- p.21 / Chapter 3.2 --- Incremental Progressive Stochastic Search --- p.23 / Chapter 3.2.1 --- Network Architecture --- p.24 / Chapter 3.2.2 --- Convergence Procedure --- p.24 / Chapter 3.2.3 --- An Illustrative Example --- p.25 / Chapter 3.3 --- Heuristic Cluster Selection Strategy --- p.28 / Chapter 4 --- Experiments --- p.31 / Chapter 4.1 --- N-Queens Problems --- p.32 / Chapter 4.2 --- Permutation Generation Problems --- p.53 / Chapter 4.2.1 --- Increasing Permutation Problems --- p.54 / Chapter 4.2.2 --- Random Permutation Generation Problems --- p.75 / Chapter 4.3 --- Latin Squares and Quasigroup Completion Problems --- p.96 / Chapter 4.3.1 --- Latin Square Problems --- p.96 / Chapter 4.3.2 --- Quasigroup Completion Problems --- p.118 / Chapter 4.4 --- Random CSPs --- p.120 / Chapter 4.4.1 --- Tight Random CSPs --- p.139 / Chapter 4.4.2 --- Phase Transition Random CSPs --- p.156 / Chapter 5 --- Concluding Remarks --- p.159 / Chapter 5.1 --- Contributions --- p.159 / Chapter 5.2 --- Future Work --- p.161

Thesis (M.S.)--Massachusetts Institute of Technology, Dept. of Political Science, 1982. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND DEWEY. / Includes bibliographies. / by Michael Cox Ryan. / M.S.

In this dissertation, I investigate how government policies influence an individual's decision to search for and accept a job and/or crime opportunity. Chapter 1 looks at how long it takes for released inmates to find a job, and when they find a job, how their incarceration rate changes. The purpose is to predict the effects of a successful job placement program. An on-the-job search model with crime is used to model criminal behavior, derive the estimation method and analyze different types of policies. The results show the unemployed are incarcerated twice as fast as the employed and take on average four and a half months to find a job. Combining these results, it is demonstrated that reducing the average unemployment spell of criminals by two months reduces crime and recidivism by more than five percent. Chapter 2 incorporates crime into a search and matching model of the labor market. All workers, irrespective of their labor force status can commit crimes and the employment contract is determined optimally. The model is used to study, analytically and quantitatively, the effects of various labor market and crime policies such as unemployment insurance, hiring subsidies and the duration of jail sentences. For example, wage subsidies reduce unemployment, the crime rates of employed and unemployed workers, and improve society's welfare. Chapter 3 investigates a market where wholesalers search for retailers and retailers search for consumers. I show how the timing, targets and types of anti-drug policies matter. For instance, supply falls if the likelihood of apprehension rises when a network is established. Alternatively, if the cost of apprehension rises for wholesale dealers when a network is searching for consumers, then revenue sharing is distorted. Such a distortion will increase retail profits and aggregate supply. As an application, the model provides an alternative explanation for why the United States cocaine market saw rising consumption and falling prices during the 1980's. Specifically, the ``War on Drugs" distorted the cocaine market and increased supply.

Information available in digital form has grown phenomenally in recent years. Finding the required information has become a difficult and challenging task. This is primarily due to the diversity and enormous volume of information available and the change in the nature of people now seeking information – from experts to ordinary users of desktop computers with varying interest and objectives. The problem of finding relevant information is further impacted by the poor retrieval effectiveness of most current information retrieval (IR) systems that are primarily based on keyword indexing techniques. Though these systems retrieve documents that contain those keywords specified in the query, the documents that are retrieved may not necessarily be in the context in which the user would have wanted them to be. This research works argues that exploiting the user’s context of the information need has the potential to improve the performance of information retrieval systems. Context can reduce the ambiguity by associating meanings to request/query terms, and thus limit the scope of the possible misinterpretations of query terms. A new way of defining context categories based on information type is proposed and this notion of context differs from the conventional way of defining information categories based on subject topics as it is closely linked with the situation in which the user’s needs for information originates. A new context-based information retrieval system where users could specify the context in which they are seeking information is presented. This work also includes a full-scale development, implementation and evaluation of the new context-based information system / Doctor of Philosophy (PhD)