In today's information age, data collection is not the ultimate goal; it is simply the first step in extracting knowledge-rich information to shape future decisions. In this thesis, we present ChartVisio - a simple web-based visual data-mining system that lets users quickly explore databases and transform raw data into processed visuals. It is highly interactive, easy to use and hides the underlying complexity of querying from its users. Data from tables is internally mapped into charts using aggregate functions across tables. The tool thus integrates querying and charting into a single general-purpose application. ChartVisio has been designed as a component of the Benchmark data engine, being developed at the Computer Science department, University of New Orleans. The data engine is an intelligent website generator and users who create websites using the Data Engine are the site owners. Using ChartVisio, owners may generate new charts and save them as XML templates for prospective website surfers. Everyday Internet users may view saved charts with the touch of a button and get real-time data, since charts are generated dynamically. Website surfers may also generate new charts, but may not save them as templates. As a result, even non-technical users can design and generate charts with minimal time and effort.

Data mining for patterns and knowledge discovery in multivariate datasets are very important processes and tasks to help analysts understand the dataset, describe the dataset, and predict unknown data values. However, conventional computer-supported data mining approaches often limit the user from getting involved in the mining process and performing interactions during the pattern discovery. Besides, without the visual representation of the extracted knowledge, the analysts can have difficulty explaining and understanding the patterns. Therefore, instead of directly applying automatic data mining techniques, it is necessary to develop appropriate techniques and visualization systems that allow users to interactively perform knowledge discovery, visually examine the patterns, adjust the parameters, and discover more interesting patterns based on their requirements. In the dissertation, I will discuss different proposed visualization systems to assist analysts in mining patterns and discovering knowledge in multivariate datasets, including the design, implementation, and the evaluation. Three types of different patterns are proposed and discussed, including trends, clusters of subgroups, and local patterns. For trend discovery, the parameter space is visualized to allow the user to visually examine the space and find where good linear patterns exist. For cluster discovery, the user is able to interactively set the query range on a target attribute, and retrieve all the sub-regions that satisfy the user's requirements. The sub-regions that satisfy the same query and are neareach other are grouped and aggregated to form clusters. For local pattern discovery, the patterns for the local sub-region with a focal point and its neighbors are computationally extracted and visually represented. To discover interesting local neighbors, the extracted local patterns are integrated and visually shown to the analysts. Evaluations of the three visualization systems using formal user studies are also performed and discussed.

Made available in DSpace on 2014-06-12T16:01:08Z (GMT). No. of bitstreams: 2 arquivo7170_1.pdf: 2203364 bytes, checksum: a4b1c6049227e992e107cabafa05f77c (MD5) license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) Previous issue date: 2005 / A descoberta de conhecimento em bases de dados (Knowledge Discovery in Databases KDD) visa a apoiar os processos de tomada de decisão através da extração automática de conhecimento oculto, útil e estratégico, em grandes bases de dados. Este conhecimento precisa ser analisado e facilmente entendido por usuários e gestores para que se torne realmente relevante nas operações cotidianas ou em planejamento de ações no contexto do problema analisado. O conhecimento descoberto pode ser apresentado de diversas formas. Entretanto, estas formas muitas vezes não são compreendidas pelo usuário ou não permitem análises detalhadas e validações de novas hipóteses. Para auxiliar a interpretação de resultados obtidos na mineração de dados, técnicas gráficas de Visualização de Informações têm contribuído significativamente para a representação inteligente de grandes volumes de dados, para a aplicação de técnicas estatísticas na análise de dados e para a manipulação visual dos dados. À aplicação dessas técnicas sobre o processo de KDD dá-se o nome de Visual Data Mining. Os principais objetivos deste trabalho são a investigação de técnicas de Visualização de Informações aplicadas no processo de KDD, o desenvolvimento de uma ferramenta de software que tenha foco principal em Visual Data Mining, com a proposição e implementação de técnicas e métodos que melhor se adaptem à interpretação de resultados minerados, e a realização de um estudo de caso com um problema em larga escala para validação da ferramenta desenvolvida. A ferramenta desenvolvida, denominada VisualDATAMINER , atua sobre a interpretação de regras de indução, permite a integração com ferramentas de mineração de dados, possibilita a visualização dos resultados de mineração de dados em diversas visões e a interação com estas visualizações através de métodos de interação. Desenvolvida na linguagem Java, a VisualDATAMINER apresenta todos os benefícios do paradigma de orientação a objetos como re-usabilidade, manutenibilidade e encapsulamento. A investigação experimental realizada usando uma base de dados com um grande volume de dados, no domínio de análise de crédito ao consumidor, mostrou o refinamento do conhecimento descoberto através da aplicação das técnicas de visualização de informações e dos métodos de interação propostos na ferramenta, atestando a eficácia e a eficiência da ferramenta desenvolvida

Deep models are state-of-the-art for many computer vision tasks including object classification, action recognition, and captioning. As Artificial Intelligence systems that utilize deep models are becoming ubiquitous, it is also becoming crucial to explain why they make certain decisions: Grounding model decisions. In this thesis, we study: 1) Improving Model Classification. We show that by utilizing web action images along with videos in training for action recognition, significant performance boosts of convolutional models can be achieved. Without explicit grounding, labeled web action images tend to contain discriminative action poses, which highlight discriminative portions of a video’s temporal progression. 2) Spatial Grounding. We visualize spatial evidence of deep model predictions using a discriminative top-down attention mechanism, called Excitation Backprop. We show how such visualizations are equally informative for correct and incorrect model predictions, and highlight the shift of focus when different training strategies are adopted. 3) Spatial Grounding for Improving Model Classification at Training Time. We propose a guided dropout regularizer for deep networks based on the evidence of a network prediction. This approach penalizes neurons that are most relevant for model prediction. By dropping such high-saliency neurons, the network is forced to learn alternative paths in order to maintain loss minimization. We demonstrate better generalization ability, an increased utilization of network neurons, and a higher resilience to network compression. 4) Spatial Grounding for Improving Model Classification at Test Time. We propose Guided Zoom, an approach that utilizes spatial grounding to make more informed predictions at test time. Guided Zoom compares the evidence used to make a preliminary decision with the evidence of correctly classified training examples to ensure evidenceprediction consistency, otherwise refines the prediction. We demonstrate accuracy gains for fine-grained classification. 5) Spatiotemporal Grounding. We devise a formulation that simultaneously grounds evidence in space and time, in a single pass, using top-down saliency. We visualize the spatiotemporal cues that contribute to a deep recurrent neural network’s classification/captioning output. Based on these spatiotemporal cues, we are able to localize segments within a video that correspond with a specific action, or phrase from a caption, without explicitly optimizing/training for these tasks.

The diploma thesis is focused on problematic of management of digital libraries, more precisely it’s focused on standard MPEG-7 ISO/IEC 15938. The thesis defines description of multimedia content, its basic rules and more deeply it describes work with static visual data. The practical part of the thesis is focused on creating standalone application in programming language Java. Its purpose is using XML files to manage, save, search and for general work with visual data and its parameters. An application is focused on Scalable Color and Dominant Color descriptors and propose alternative calculation procedure by variable parameters.

International Telemetering Conference Proceedings / October 23-26, 2000 / Town & Country Hotel and Conference Center, San Diego, California / With wireless communications becoming the rule rather than the exception, satellite operators need tools to effectively monitor increasingly large and complex satellite constellations. Visual data monitoring increases the monitoring capacity of satellite operators by several orders of magnitude, enabling them to track hundreds of thousands of parameters in real-time on a single screen. With this powerful new tool, operators can proactively address potential problems before they become customer complaints.

Data stream management and data flow programming have many things in common. In both cases one wants to transfer possibly infinite sequences of data items from one place to another, while performing transformations to the data. This Thesis focuses on the integration of a visual programming language with a data stream management system (DSMS) to support the construction, configuration, and visualization of data stream applications. In the approach, analyses of data streams are expressed as continuous queries (CQs) that emit data in real-time. The LabVIEW visual programming platform has been adapted to support easy specification of continuous visualization of CQ results. LabVIEW has been integrated with the DSMS SVALI through a stream-oriented client-server API. Query programming is declarative, and it is desirable to make the stream visualization declarative as well, in order to raise the abstraction level and make programming more intuitive. This has been achieved by adding a set of visual data flow components (VDFCs) to LabVIEW, based on the LabVIEW actor framework. With actor-based data flows, visualization of data stream output becomes more manageable, avoiding the procedural control structures used in conventional LabVIEW programming while still utilizing the comprehensive, built-in LabVIEW visualization tools. The VDFCs are part of the Visual Data stream Monitor (VisDM), which is a client-server based platform for handling real-time data stream applications and visualizing stream output. VDFCs are based on a data flow framework that is constructed from the actor framework, and are divided into producers, operators, consumers, and controls. They allow a user to set up the interface environment, customize the visualization, and convert the streaming data to a format suitable for visualization. Furthermore, it is shown how LabVIEW can be used to graphically define interfaces to data streams and dynamically load them in SVALI through a general wrapper handler. As an illustration, an interface has been defined in LabVIEW for accessing data streams from a digital 3D antenna. VisDM has successfully been tested in two real-world applications, one at Sandvik Coromant and one at the Ångström Laboratory, Uppsala University. For the first case, VisDM was deployed as a portable system to provide direct visualization of machining data streams. The data streams can differ in many ways as do the various visualization tasks. For the second case, data streams are homogenous, high-rate, and query operations are much more computation-demanding. For both applications, data is visualized in real-time, and VisDM is capable of sufficiently high update frequencies for processing and visualizing the streaming data without obstructions. The uniqueness of VisDM is the combination of a powerful and versatile DSMS with visually programmed and completely customizable visualization, while maintaining the complete extensibility of both.

Há várias décadas, pesquisadores em ciências sociais buscam formas gráficas para expressar as relações humanas na sociedade. O advento do computador e, mais recentemente, da internet, possibilitou o surgimento de um campo que tem despertado a atenção de estudiosos das áreas de visualização de informação e de ciências sociais, o da visualização de redes sociais. Esse campo tem o potencial de revelar e explorar padrões que podem beneficiar um número muito grande de aplicações e indivíduos em áreas tais como comércio, segurança em geral, redes de conhecimento e pesquisa de mercado. Grande parte dos algoritmos de visualização de redes sociais são baseados em grafos, destacando relacionamentos entre indivíduos e grupos de indivíduos, mas dando pouca atenção aos seus demais atributos. Assim, este trabalho apresenta um conjunto de soluções para representar e explorar visualmente redes sociais levando em consideração tais atributos. A primeira solução faz uso de redes heterogêneas, onde tanto indivíduos quanto comunidades são representados no grafo; a segunda solução utiliza técnicas de visualização baseadas em projeção multidimensional, que promovem o posicionamento dos dados no plano de acordo com algum critério de similaridade baseado em atributo; e a última solução coordena múltiplas visões para focar rapidamente em regiões de interesse. Os resultados indicam que as soluções proveem um poder de representação e identificação de conceitos não facilmente detectados por formas convencionais de visualização e exploração de grafos, com indícios fornecidos através dos estudos de caso e da realização de avaliações com usuários. Este trabalho fornece um estudo das áreas de visualização em grafos para a análise de redes sociais bem como uma implementação das soluções de integração da visualização em redes com as projeções multidimensionais / For decades, social sciences researchers have searched for graphical forms to express human social relationships. The development of computer science and more recently of the Internet has given rise to a new field of research for visualization and social sciences professionals, that of social network visualization. This field can potentially offer new opportunities in reveal new patterns that can benefit a large number of applications and individuals in fields such as commerce, security, knowledge networks and marketing. A large part of social network visualization algorithms and systems relies on graph representations, highlighting relationships amongst individuals and groups of individuals, but mostly neglecting the other available attributes of individuals. Thus, this work presents a set of tools to represent and explore social networks visually, taking into consideration the attributes of the nodes. The first technique employs heterogeneous networks, where both individuals and communities are represented in the graph; the second solution uses visualization techniques based on multidimensional projection, which promote the placement of data in the plane according to some similarity criterion based on attribute; still another proposed technique coordinates multiple views in order to speed up focus in regions of interest in the data sets. The results indicate that the solutions promote high degree of representation power and that concept identification not easily obtained via other methods is possible; the evidence comes from case studies as well as a user evaluation. This work includes a study in the area of graph visualization for social network analysis as well as a system implementing the proposed solutions, that integrate network visualization and multidimensional projections to extract patterns from social networks

Automatic analysis of visual data is a key objective of computer vision research; and performing visual recognition of objects from images is one of the most important steps towards understanding and gaining insights into the visual data. Most existing approaches in the literature for the visual recognition are based on a supervised learning paradigm. Unfortunately, they require a large amount of labelled training data which severely limits their scalability. On the other hand, recognition is instantaneous and effortless for humans. They can recognise a new object without seeing any visual samples by just knowing the description of it, leveraging similarities between the description of the new object and previously learned concepts. Motivated by humans recognition ability, this thesis proposes novel approaches to tackle cross-class transfer learning (crossclass recognition) problem whose goal is to learn a model from seen classes (those with labelled training samples) that can generalise to unseen classes (those with labelled testing samples) without any training data i.e., seen and unseen classes are disjoint. Specifically, the thesis studies and develops new methods for addressing three variants of the cross-class transfer learning: Chapter 3 The first variant is transductive cross-class transfer learning, meaning labelled training set and unlabelled test set are available for model learning. Considering training set as the source domain and test set as the target domain, a typical cross-class transfer learning assumes that the source and target domains share a common semantic space, where visual feature vector extracted from an image can be embedded using an embedding function. Existing approaches learn this function from the source domain and apply it without adaptation to the target one. They are therefore prone to the domain shift problem i.e., the embedding function is only concerned with predicting the training seen class semantic representation in the learning stage during learning, when applied to the test data it may underperform. In this thesis, a novel cross-class transfer learning (CCTL) method is proposed based on unsupervised domain adaptation. Specifically, a novel regularised dictionary learning framework is formulated by which the target class labels are used to regularise the learned target domain embeddings thus effectively overcoming the projection domain shift problem. Chapter 4 The second variant is inductive cross-class transfer learning, that is, only training set is assumed to be available during model learning, resulting in a harder challenge compared to the previous one. Nevertheless, this setting reflects a real-world setting in which test data is available after the model learning. The main problem remains the same as the previous variant, that is, the domain shift problem occurs when the model learned only from the training set is applied to the test set without adaptation. In this thesis, a semantic autoencoder (SAE) is proposed building on an encoder-decoder paradigm. Specifically, first a semantic space is defined so that knowledge transfer is possible from the seen classes to the unseen classes. Then, an encoder aims to embed/project a visual feature vector into the semantic space. However, the decoder exerts a generative task, that is, the projection must be able to reconstruct the original visual features. The generative task forces the encoder to preserve richer information, thus the learned encoder from seen classes is able generalise better to the new unseen classes. Chapter 5 The third one is unsupervised cross-class transfer learning. In this variant, no supervision is available for model learning i.e., only unlabelled training data is available, leading to the hardest setting compared to the previous cases. The goal, however, is the same, learning some knowledge from the training data that can be transferred to the test data composed of completely different labels from that of training data. The thesis proposes a novel approach which requires no labelled training data yet is able to capture discriminative information. The proposed model is based on a new graph regularised dictionary learning algorithm. By introducing a l1- norm graph regularisation term, instead of the conventional squared l2-norm, the model is robust against outliers and noises typical in visual data. Importantly, the graph and representation are learned jointly, resulting in further alleviation of the effects of data outliers. As an application, person re-identification is considered for this variant in this thesis.

By increasing the volume and complexity of datasets, Visual Data Mining (VDM), new visualization techniques evolved and new techniques released. However, some of these techniques performing well and cover all expectations; the others failed to save their positions. The main issue of such techniques is problem dependency.In this study, after a short description about necessity of Visual Data Mining techniques, I will provide a classified review of previous researches. This will result in a deep understanding as well as simple accessibility to previous researches, in a concise manner. This will facilitate the extraction of the specifications of 3D visualization technique and will provide a comprehensive knowledge of this technique in a classified manner. After that, all possible combination of 3D visualization technique will review.3D Visualization technique as a popular technique is a concrete foundation for visualization of multi-dimensional datasets, but it has some limitations. To overcome these limitations, previous studies in literature as well as the experiences of professionals will gather. The results will prove the theoretical findings as well as offering new hybrid techniques (combination with 3D visualization and other visual data mining techniques).The contribution of professionals will empower and complement the results of this study, as they can address solutions for the weaknesses of 3D Visualization technique in their business which is new combination of techniques. These combinations of techniques will create the basis for future researches in order to discover new limitations and provide solutions to overcome by use of hybrid techniques. / Program: Magisterutbildning i informatik