No description available.

Beamfilling error is one of the main error sources for microwave oceanic rainfall retrieval. An accurate beamfilling correction can improve the rainfall retrieval accuracy significantly. Quantitative understanding of the uncertainty of the Beamfilling Correction Factor (BCF) is very important for the understanding of the accuracy of microwave passive rainfall retrieval. Refinement of the calculation of the BCF and the estimation of BCF uncertainty are the main purposes of this thesis. The characteristic of rainfall distribution is investigated. Quantitative understanding of the statistical characteristics of rainfall distribution provides an indication of the beamfilling error and the uncertainty of BCF in many ways. Some refinements to the traditional BCF calculation algorithm are provided in this thesis. Scattering is included in the new algorithm. Also the BCF calculation only considers the cases within the useful dynamic range. These refinements make the BCF calculation closer to how it is used in the retrieval algorithm. The BCF based on the new algorithm should be more accurate. The global BCF uncertainty and the local BCF uncertainty are estimated using the available A/C radar data. The results show that the uncertainty of BCF is much smaller than expected, and also show that the BCF derived from a specific set of data can be used globally.

Traditional information retrieval applications, such as Web search, return atomic units of retrieval, which are generically called ``documents''. Depending on the application, a document may be a Web page, an email message, a journal article, or any similar object. In contrast to this traditional approach, focused retrieval helps users better pin-point their exact information needs by returning results at the sub-document level. These results may consist of predefined document components~---~such as pages, sections, and paragraphs~---~or they may consist of arbitrary passages, comprising any sub-string of a document. If a document is marked up with XML, a focused retrieval system might return individual XML elements or ranges of elements. This thesis proposes and evaluates a number of approaches to focused retrieval, including methods based on XML markup and methods based on arbitrary passages. It considers the best unit of retrieval, explores methods for efficient sub-document retrieval, and evaluates formulae for sub-document scoring. Focused retrieval is also considered in the specific context of the Wikipedia, where methods for automatic vandalism detection and automatic link generation are developed and evaluated.

One of the greatest challenges in information retrieval is to develop an intelligent system for user and machine interaction that supports users in their quest for relevant information. The dramatic increase in the amount of Web content gives rise to the need for a large-scale distributed information retrieval system, targeted to support millions of users and terabytes of data. To retrieve information from such a large amount of data in an efficient manner, the index is split among the servers in a distributed information retrieval system. Thus, partitioning the index among these collaborating nodes plays an important role in enhancing the performance of a distributed search engine. The two widely known inverted index partitioning schemes for a distributed information retrieval system are document partitioning and term partitioning. %In a document partitioned system, each of the server hosts a subset of the documents in the collection, and execute every query against its local sub-collection. In a term partitioned index, each node is responsible for a subset of the terms in the collection, and serves them to a central node as they are required for query evaluation. In this thesis, we introduce the Document over Term inverted index distribution scheme, which splits a set of nodes into several groups (sub-clusters) and then performs document partitioning between the groups and term partitioning within the group. As this approach is based on the term and document index partitioning approaches, we also refer it as a Hybrid Inverted Index. This approach retains the disk access benefits of term partitioning and the benefits of sharing computational load, scalability, maintainability, and availability of the document partitioning. We also introduce the Document over Document index partitioning scheme, based on the document partitioning approach. In this approach, a set of nodes is split into groups and documents in the collection are partitioned between groups and also within each group. This strategy retains all the benefits of the document partitioning approach, but reduces the computational load more effectively and uses resources more efficiently. We compare distributed index approaches experimentally and show that in terms of efficiency and scalability, document partition based approaches perform significantly better than the others. The Document over Term partitioning offers efficient utilization of search-servers and lowers disk access, but suffers from the problem of load imbalance. The Document over Document partitioning emerged to be the preferred method during high workload.

No description available.

No description available.

No description available.

No description available.

No description available.

As a large volume of digital video data becomes available, along with revolutionary advances in multimedia technologies, demand related to efficiently retrieving and recounting multimedia data has grown. However, the inherent complexity in representing and recognizing multimedia data, especially for large-scale and unconstrained consumer videos, poses significant challenges. In particular, the following challenges are major concerns in the proposed research. One challenge is that consumer-video data (e.g., videos on YouTube) are mostly unstructured; therefore, evidence for a targeted semantic category is often sparsely located across time. To address the issue, a segmental multi-way local feature pooling method by using scene concept analysis is proposed. In particular, the proposed method utilizes scene concepts that are pre-constructed by clustering video segments into categories in an unsupervised manner. Then, a video is represented with multiple feature descriptors with respect to scene concepts. Finally, multiple kernels are constructed from the feature descriptors, and then, are combined into a final kernel that improves the discriminative power for multimedia event detection. Another challenge is that most semantic categories used for multimedia retrieval have inherent within-class diversity that can be dramatic and can raise the question as to whether conventional approaches are still successful and scalable. To consider such huge variability and further improve recounting capabilities, a per-exemplar learning scheme is proposed with a focus on fusing multiple types of heterogeneous features for video retrieval. While the conventional approach for multimedia retrieval involves learning a single classifier per category, the proposed scheme learns multiple detection models, one for each training exemplar. In particular, a local distance function is defined as a linear combination of element distance measured by each features. Then, a weight vector of the local distance function is learned in a discriminative learning method by taking only neighboring samples around an exemplar as training samples. In this way, a retrieval problem is redefined as an association problem, i.e., test samples are retrieved by association-based rules. In addition, the quality of a multimedia-retrieval system is often evaluated by domain-specific performance metrics that serve sophisticated user needs. To address such criteria for evaluating a multimedia-retrieval system, in MFoM learning, novel algorithms were proposed to explicitly optimize two challenging metrics, AP and a weighted sum of the probabilities of false alarms and missed detections at a target error ratio. Most conventional learning schemes attempt to optimize their own learning criteria, as opposed to domain-specific performance measures. By addressing this discrepancy, the proposed learning scheme approximates the given performance measure, which is discrete and makes it difficult to apply conventional optimization schemes, with a continuous and differentiable loss function which can be directly optimized. Then, a GPD algorithm is applied to optimizing this loss function.