Proceedings of the 26th International Conference on Data Engineering



Search and Social Integration.

AdHeat an influence%2Dbased diffusion model for propagating hints to match ads.
In this paper, we present AdHeat, a social ad model considering user influence in addition to relevance for matching ads. Traditionally, ad placement employs the relevance model. Such a model matches ads with Web page content, user interests, or both. We have observed, however, on social networks that the relevance model suffers from two shortcomings. First, influential users (users who contribute opinions) seldom click ads that are highly relevant to their expertise. Second, because influential users ’ contents and activities are attractive to other users, hint words summarizing their expertise and activities may be widely preferred. Therefore, we propose AdHeat, which diffuses hint words of influential users to others and then matches ads for each user with aggregated hints. We performed experiments on a large online Q&amp;A community with half a million users. The experimental results show that AdHeat outperforms the relevance model on CTR (click through rate) by significant margins. Categories and Subject Descriptors
Confucius and Its Intelligent Disciples Integrating Social with Search.
Q&amp;A sites continue to flourish as a large number of users rely on them as useful substitutes for incomplete or missing search results. In this paper, we present our experience with developing Confucius, a Google Q&amp;A service launched in 21 countries and four languages by the end of 2009. Confucius employs six data mining subroutines to harness synergy between web search and social networks. We present these subroutines ’ design goals, algorithms, and their effects on service quality. We also describe techniques for and experience with scaling the subroutines to mine massive data sets. 1.
Parallel Algorithms for Collaborative Filtering.
Workshop Program
Parallel algorithms for mining large%2Dscale rich%2Dmedia data.

PLDA Parallel Latent Dirichlet Allocation for Large%2DScale Applications.
Abstract. This paper presents PLDA, our parallel implementation of Latent Dirichlet Allocation on MPI and MapReduce. PLDA smooths out storage and computation bottlenecks and provides fault recovery for lengthy distributed computations. We show that PLDA can be applied to large, real-world applications and achieves good scalability. We have released MPI-PLDA to open source at http://code.google.com/p/plda under the Apache License. 1
Parallel algorithms for mining large%2Dscale rich%2Dmedia data.

Confucius and %22Its%22 Intelli

Confucius and %22Its%22 Intelli

Confucius and %22its%22 intelligent disciples.
Q&amp;A sites continue to flourish as a large number of users rely on them as useful substitutes for incomplete or missing search results. In this paper, we present our experience with developing Confucius, a Google Q&amp;A service launched in 21 countries and four languages by the end of 2009. Confucius employs six data mining subroutines to harness synergy between web search and social networks. We present these subroutines ’ design goals, algorithms, and their effects on service quality. We also describe techniques for and experience with scaling the subroutines to mine massive data sets. 1.
Confucius and %22its%22 intelligent disciples.
Q&amp;A sites continue to flourish as a large number of users rely on them as useful substitutes for incomplete or missing search results. In this paper, we present our experience with developing Confucius, a Google Q&amp;A service launched in 21 countries and four languages by the end of 2009. Confucius employs six data mining subroutines to harness synergy between web search and social networks. We present these subroutines ’ design goals, algorithms, and their effects on service quality. We also describe techniques for and experience with scaling the subroutines to mine massive data sets. 1.
Information extraction meets relation databases.

Information extraction meets relation databases.



Semantic Modeling and Knowledge Representation for Multimedia Data.

Organizing multimedia data socially.

Parallel Spectral Clustering.
Abstract. Spectral clustering algorithm has been shown to be more effective in finding clusters than most traditional algorithms. However, spectral clustering suffers from a scalability problem in both memory use and computational time when a dataset size is large. To perform clustering on large datasets, we propose to parallelize both memory use and computation on distributed computers. Through an empirical study on a large document dataset of 193, 844 data instances and a large photo dataset of 637, 137, we demonstrate that our parallel algorithm can effectively alleviate the scalability problem. Key words: Parallel spectral clustering, distributed computing 1
Combinational collaborative filtering for personalized community recommendation.
Rapid growth in the amount of data available on social networking sites has made information retrieval increasingly challenging for users. In this paper, we propose a collaborative filtering method, Combinational Collaborative Filtering (CCF), to perform personalized community recommendations by considering multiple types of co-occurrences in social data at the same time. This filtering method fuses semantic and user information, then applies a hybrid training strategy that combines Gibbs sampling and Expectation-Maximization algorithm. To handle the large-scale dataset, parallel computing is used to speed up the model training. Through an empirical study on the Orkut dataset, we show CCF to be both effective and scalable.

Web%2DScale Image Annotation.
Abstract. In this paper, we describe our experiments using Latent Dirichlet Allocation (LDA) to model images containing both perceptual features and words. To build a large-scale image tagging system, we distribute the computation of LDA parameters using MapReduce. Empirical study shows that our scalable LDA supports image annotation both effectively and efficiently.

Rich media and web 2.0.

Scalable landmark recognition using EXTENT.

Data management projects at Google.

Bayesian Reasoning for Sensor Group%2DQueries and Diagnosis.

Parallel Approximate Matrix Factorization for Kernel Methods.

Scalable Indexing for Perceptual Data.

Parallelizing Support Vector Machines on Distributed Computers.
Support Vector Machines (SVMs) suffer from a widely recognized scalability problem in both memory use and computational time. To improve scalability, we have developed a parallel SVM algorithm (PSVM), which reduces memory use through performing a row-based, approximate matrix factorization, and which loads only essential data to each machine to perform parallel computation. Let n denote the number of training instances, p the reduced matrix dimension after factorization (p is significantly smaller than n), and m the number of machines. PSVM reduces the memory requirement from O(n 2) to O(np/m), and improves computation time to O(np 2 /m). Empirical study shows PSVM to be effective. PSVM Open Source is available for download at
RCMap Efficiently Creating High%2DQuality Euclidean Embeddings.

Semiparametric Regression Using Student t Processes.

Building MEMS%2Dbased storage systems for streaming media.
The performance of streaming media servers has been limited by the dual requirements of high disk throughput (to service more clients simultaneously) and low memory use (to decrease system cost). To achieve high disk throughput, disk drives must be accessed with large IOs to amortize disk access overhead. Large IOs imply an increased requirement of expensive DRAM, and, consequently, greater overall system cost. MEMS-based storage, an emerging storage technology, is predicted to offer a price-performance point between those of DRAM and disk drives. In this study, we propose storage architectures that use the relatively inexpensive MEMS-based storage devices as an intermediate layer (between DRAM and disk drives) for temporarily staging large disk IOs at a significantly lower cost. We present data layout mechanisms and synchronized IO scheduling algorithms for the real-time storage and retrieval of streaming data within such an augmented storage system. Analytical evaluation suggests that MEMS-augmented storage hierarchies can reduce the cost and improve the throughput of streaming servers significantly. Categories and Subject Descriptors: C.0.a [Computer System Organization]: General—System
Efficient top%2Dk hyperplane query processing for multimedia information retrieval.
A query can be answered by a binary classifier, which separates the instances that are relevant to the query from the ones that are not. When kernel methods are employed to train such a classifier, the class boundary is represented as a hyperplane in a projected space. Data instances that are farthest from the hyperplane are deemed to be most relevant to the query, and that are nearest to the hyperplane to be most uncertain to the query. In this paper, we address the twin problems of efficient retrieval of the approximate set of instances (a) farthest from and (b) nearest to a query hyperplane. Retrieval of instances for this hyperplane-based query scenario is mapped to the range-query problem allowing for the reuse of existing index structures. Empirical evaluation on large image datasets confirms the effectiveness of our approach.
Fotofiti web service for photo management.
In this work, we present Fotofiti(FF), a web-based personal photo organizer with automatic image annotation, event management and social network integration. We describe our technique for real-time online semantic annotation of user photos. Additionally, a landmark recognition system which utilizes local features is discussed.
A scalable service for photo annotation%2C sharing%2C and search.
In this work we present the details of the implementation of Fotofiti(FF), a website that provides automatic semantic annotation of digital photographs, event management and social network integration. We describe our technique for real-time online semantic annotation using global features from both content and context. Classification experiments using various learning techniques were performed on a realworld data-set. Additionally, a scalable landmark recognition system which utilizes local features is discussed.
Fotowiki distributed map enhancement service.

Adaptive non%2Dlinear clustering in data streams.

Mining blog stories using community%2Dbased and temporal clustering.

Identifying Color in Motion in Video Sensors.
Identifying or matching the surface color of a moving object in surveillance video is critical for achieving reliable object-tracking and searching. Traditional color models provide little help, since the surface of an object is usually not flat, the object’s motion can alter the surface’s orientation, and the lighting conditions can vary when the object moves. To tackle this research problem, we conduct extensive data mining on video clips collected under various lighting conditions and distances from several video-cameras. We observe how each of the eleven culture colors can drift in the color space when an object’s surface is in motion. In the color space, we then learn the drift pattern of each culture color for classifying unseen surface colors. Finally, we devise a distance function taking color drift into consideration to perform color identification and matching. Empirical studies show our approach to be very promising: achieving over 95 % color-prediction accuracy. 1
Hypersphere Indexer.
Abstract ⎯ Feature selection has been used widely for a variety of data, yielding higher speeds and reduced computational cost for the classification process. However, it is in microarray datasets where its advantages become more evident and are more required. In this paper we present a novel approach to accomplish this based on the concept of discernibility that we introduce to depict how separated the classes of a dataset are. We develop and test two independent feature selection methods that follow this approach. The results of our experiments on four microarray datasets show that discernibility-based feature selection reduces the dimensionality of the datasets involved without compromising the performance of the classifiers. Index Terms⎯classification problems, feature selection, dimensionality reduction, discernibility, microarray data. I.
Concept boundary detection for speeding up SVMs.
Support Vector Machines (SVMs) suffer from an O(n 2) training cost, where n denotes the number of training instances. In this paper, we propose an algorithm to select boundary instances as training data to substantially reduce n. Our proposed algorithm is motivated by the result of (Burges, 1999) that, removing non-support vectors from the training set does not change SVM training results. Our algorithm eliminates instances that are likely to be non-support vectors. In the conceptindependent preprocessing step of our algorithm, we prepare nearest-neighbor lists for training instances. In the concept-specific sampling step, we can then effectively select useful training data for each target concept. Empirical studies show our algorithm to be effective in reducing n, outperforming other competing downsampling algorithms without significantly compromising testing accuracy. 1.
Learning the unified kernel machines for classification.

Incremental approximate matrix factorization for speeding up support vector machines.

Unified and scalable learning in multimedia information retrieval.

KDX An Indexer for Support Vector Machines.

Active learning in very large databases.
Abstract. Query-by-example and query-by-keyword both suffer from the problem of “aliasing,” meaning that example-images and keywords potentially have variable interpretations or multiple semantics. For discerning which semantic is appropriate for a given query, we have established that combining active learning with kernel methods is a very effective approach. In this work, we first examine active-learning strategies, and then focus on addressing the challenges of two scalability issues: scalability in concept complexity and in dataset size. We present remedies, explain limitations, and discuss future directions that research might take.
Storage and Retrieval Methods and Applications for Multimedia 2005
Classifying the identities of people appearing in broadcast news video into anchor, reporter, or news subject is an important topic in high-level video analysis. Given the visual resemblance of different types of people, this work explores multi-modal features derived from a variety of evidences, such as the speech identity, transcript clues, temporal video structure, named entities, and uses a statistical learning approach to combine all the features for person type classification. Experiments conducted on ABC World News Tonight video have demonstrated the effectiveness of the approach, and the contributions of different categories of features have been compared. Keywords: Multi-modality analysis, person type classification, broadcast news video 1.
Learning with non%2Dmetric proximity matrices.

Formulating context%2Ddependent similarity functions.
Tasks of information retrieval depend on a good distance function for measuring similarity between data instances. The most effective distance function must be formulated in a context-dependent (also application-, data-, and user-dependent) way. In this paper, we present a novel method, which learns a distance function by capturing the nonlinear relationships among contextual information provided by the application, data, or user. We show that through a process called the “kernel trick, ” such nonlinear relationships can be learned efficiently in a projected space. In addition to using the kernel trick, we propose two algorithms to further enhance efficiency and effectiveness of function learning. For efficiency, we propose a SMO-like solver to achieve O(N 2) learning performance. For effectiveness, we propose using unsupervised learning in an innovative way to address the challenge of lack of labeled data (contextual information). Theoretically, we substantiate that our method is both sound and optimal. Empirically, we demonstrate that our method is effective and useful.
Multimodal metadata fusion using causal strength.

EXTENT fusing context%2C content%2C and semantic ontology for photo annotation.

Using pivots to index for support vector machine queries.

Manifold learning%2C a promised land or work in progress%3F
Tasks of image clustering and classification often deal with data of very high dimensions. To alleviate the dimensionality curse, several methods, such as Isomap, LLE and KPCA, have recently been proposed and applied to learn low-dimensional non-linear embedded manifolds in high-dimensional spaces. Unfortunately, the scenarios in which these methods appear to be effective are very contrived. In this work, we empirically examine these methods on a realistic but not-so-difficult dataset. We discuss the promises and limitations of these dimensionreduction schemes. 1.
Extent Inferring Image Metadata from Context and Content.

Formulating distance functions via the kernel trick.
Tasks of data mining and information retrieval depend on a good distance function for measuring similarity between data instances. The most effective distance function must be formulated in a contextdependent (also application-, data-, and user-dependent) way. In this paper, we propose to learn a distance function by capturing the nonlinear relationships among contextual information provided by the application, data, or user. We show that through a process called the “kernel trick, ” such nonlinear relationships can be learned efficiently in a projected space. Theoretically, we substantiate that our method is both sound and optimal. Empirically, using several datasets and applications, we demonstrate that our method is effective and useful.
Exploiting Geometry for Support Vector Machine Indexing.

Kronecker Factorization for Speeding up Kernel Machines.
In kernel machines, such as kernel principal component analysis (KPCA), Gaussian Processes (GPs), and Support Vector Machines (SVMs), the computational complexity of finding a solution is O(n 3), where n is the number of training instances. To reduce this expensive computational complexity, we propose using Kronecker factorization, which approximates a positive definite kernel matrix by the Kronecker product of two smaller positive definite matrices. This approximation can speed up the calculation of the kernel-matrix inverse or eigendecomposition involved in kernel machines. When the two factorized matrices have about the same dimensions, the computational complexity is improved from O(n 3) to O(n 2). Furthermore, if n is very large, Kronecker factorization can be recursively applied to further reduce the computational complexity. We propose two methods to carry out Kronecker factorization and apply them to speed up KPCA and GPs. In addition, we propose an effective approximate method for Gaussian process classification by integrating the surrogate maximization algorithm and the Kronecker factorization. Experiments show that our methods can drastically reduce the computation time of kernel machines without any significant degradation in their effectiveness. 1
Systems Support for Preemptive Disk Scheduling.
Abstract—Allowing higher-priority requests to preempt ongoing disk IOs is of particular benefit to delay-sensitive and real-time systems. In this paper, we present Semi-preemptible IO, which divides disk IO requests into small temporal units of disk commands to improve the preemptibility of disk access. We first lay out main design strategies to allow preemption of each component of a disk access—seek, rotation, and data transfer, namely, seek-splitting, JIT-seek, and chunking. We then present the preemption mechanisms for single and multidisk systems—JIT-preemption and JIT-migration. The evaluation of our prototype system showed that Semi-preemptible IO substantially improved the preemptibility of disk access with little loss in disk throughput and that preemptive disk scheduling could improve the response time for high-priority interactive requests. Index Terms—Storage, preemptible disk access, preemptive disk scheduling, real-time, QoS, disk IO preemption. 1
Using One%2DClass and Two%2DClass SVMs for Multiclass Image Annotation.

KBA Kernel Boundary Alignment Considering Imbalanced Data Distribution.



Semantics and feature discovery via confidence%2Dbased ensemble.


Multimodal concept%2Ddependent active learning for image retrieval.


Distance%2Dfunction design and fusion for sequence data.

On Scalability of Active Learning for Formulating Query Concepts.

Adaptive sampling for sensor networks.
A distributed data-stream architecture finds application in sensor networks for monitoring environment and activities. In such a network, large numbers of sensors deliver continuous data to a central server. The rate at which the data is sampled at each sensor affects the communication resource and the computational load at the central server. In this paper, we propose a novel adaptive sampling technique where the sampling rate at each sensor adapts to the streaming-data characteristics. Our approach employs a Kalman-Filter (KF)-based estimation technique wherein the sensor can use the KF estimation error to adaptively adjust its sampling rate within a given range, autonomously. When the desired sampling rate violates the range, a new sampling rate is requested from the server. The server allocates new sampling rates under the constraint of available resources such that KF estimation error over all the active streaming sensors is minimized. Through empirical studies, we demonstrate the flexibility and effectiveness of our model. 1
Aligning Boundary in Kernel Space for Learning Imbalanced Dataset.

Multimodal information fusion for video concept detection.

Toward building a robust and intelligent video surveillance system a case study.
With the proliferation of inexpensive cameras, availability of large-capacity disk storages, and ubiquitous presence of high-speed, broad-band communication networks, it is now economically and technically feasible to construct and deploy multi-camera video surveillance systems. In line with this is the need of intelligent, robust, (semi-)automated video analysis paradigms to assist the operators in scene analysis and event classification. In this paper, we summarize our current research work toward realizing such a multi-camera video surveillance system. 1
Active learning and its scalability for image retrieval.


Adaptive Stream Resource Management Using Kalman Filters.
A. /! /)6B$C&amp;&gt;;&lt;/ /,*6DE .!94,7*:43;&lt;/= 21210-46560   C# /N.%% ,*6DE .!94,7*:43;&lt;/= 21210-46560   )$	V.R#@.)6/	W$6C4)6&amp;&gt;&amp;-/ XK# /= 21210-46560   &amp;2C&amp;F7Z:F.% )6&amp;&gt;&amp;-/ XK# /= 21210-46560   &gt;&amp;-+\C&amp;F7Z:]?.%._^A..3 $`6,a.% ._^A..3 17 &#034;6D/0!V&amp;( (= $/b4&lt;`@ = 12480-40280  ..3 17260-41320  =  = $/[%&#034;D6&#034;4+&gt;	WCC44/%6=;0.+6,%6U4C  21210-46560    8940-39240  +&gt;	WCC44/%6=;0.+6,%6U4C  21210-46560   *,&amp;&gt;Q ^)6=&amp;&gt;=  YgfV0 17710-37150  )6=&amp;&gt;= 15320  =@L:%$&lt;bhJfV 0bWh&amp;2 JfV 11700-36090  20-37150  C  21 %% &lt; 28950-36090  20-37150  C  21210-46560   lN(7%QC4(m2 &amp;-nV$ 0-&gt;[*,5_=J6UD6D?$C&lt;bfVo; 0 &gt;[*, $g^g _=J6UD6D?$C&lt;bfVo; 29050-34000 &amp;([nV$ 0 43KC4D@ $ e=6D?(E&lt;/   6D?$C&lt;bfVo; 29050-34000  21210-46560 /F)/!J6[$&amp;-`&lt;/ &lt;bfVo; 29050-34000  21210-46560   7Z:( .7* f .h?^&amp;() $&amp;-`&lt;/ &lt;bfVo; 29050-34000  21210-46560   &amp;-C4+68_^h.D ) $&amp;-`&lt;/ &lt;bfVo; 29050-34000  21210-46560   A/3V&amp;&gt;C4+6AM^J.D &amp;-`&lt;/ &lt;bfVo; 29050-34000  21210-46560   6=N. D@ Y]...
Introduction to the special issue on video surveillance.


Confidence%2Dbased dynamic ensemble for image annotation and semantics discovery.
Providing accurate and scalable solutions to map low-level perceptual features to high-level semantics is critical for multimedia information organization and retrieval. In this paper, we propose a confidence-based dynamic ensemble (CDE) to overcome the shortcomings of the traditional static classifiers. In contrast to the traditional models, CDE can make dynamic adjustments to accommodate new semantics, to assist the discovery of useful low-level features, and to improve class-prediction accuracy. We depict two key components of CDE: a multi-level function that asserts class-prediction confidence, and the dynamic ensemble method based upon the confidence function. Through theoretical analysis and empirical study, we demonstrate that CDE is e#ective in annotating large-scale, real-world image datasets.
Multi%2Dcamera spatio%2Dtemporal fusion and biased sequence%2Ddata learning for security surveillance.
In this paper, we propose a framework for multi-camera video surveillance.  Our framework addresses the detection, represenation, and recognition of motion events. The detection phase handles spatiotemporal data fusion for efficiently and reliably extracting motion trajectories from video. The representation phase summarizes raw trajectory data to construct a hierarchical, invariant, and content-rich representation of the motion events. Finally, the recognition phase deals with learning using imbalanced training datasets and infinitedimensional data that also exhibit temporal ordering. Due to space limit, only the following two components are discussed in the paper: fusing spatio-temporal information from multiple camera sources and characterizing and detecting suspicious surveillance events.
Invariant Feature Extraction and Biased Statistical Inference for Video Surveillance.
Using cameras for detecting hazardous or suspicious events has spurred new research for security concerns. To make such detection reliable, researchers must overcome difficulties such as variation in camera capabilities, environmental factors, imbalances of positive and negative training data, and asymmetric  costs of misclassifying events of different classes. Following  up on the event-detection framework that we proposed in [12], we present in this paper the framework&#039;s two major components:  invariant feature extraction and biased statistical inference.
Enhancing DPF for Near%2Dreplica Image Recognition.

Design and Implementation of Semi%2Dpreemptible IO.
Allowing higher-priority requests to preempt ongoing disk IOs is of particular benefit to delay-sensitive multimedia and real-time systems. In this paper we propose Semi-preemptible IO, which divides an IO request into small temporal units of disk commands to enable preemptible disk access. We present main design strategies to allow preemption of each component of a disk access---seek, rotation, and data transfer. We analyze the performance and describe implementation challenges. Our evaluation shows that Semi-preemptible IO can substantially reduce IO waiting time with little loss in disk throughput. For example, expected waiting time for disk IOs in a video streaming system is reduced 2.1 times with the throughput loss of less than 6 percent.
MEMS%2Dbased Disk Buffer for Streaming Media Servers.
The performance of streaming media servers has been limited due to the dual requirements of high throughput and low memory use. Although disk throughput has been enjoying a 40% annual increase, slower improvements in disk access times necessitate the use of large DRAM buffers to improve the overall streaming throughput. MEMS-based storage is an exciting new technology that promises to bridge the widening performance gap between DRAM and disk-drives in the memory hierarchy. This paper explores the impact of integrating these devices into the memory hierarchy on the class of streaming media applications. We evaluate the use of MEMS-based storage for buffering and caching streaming data. We also show how a bank of k MEMS devices can be managed in either configuration and that they can provide a k-fold improvement in both throughput and access latency. An extensive analytical study shows that using MEMS storage can reduce the buffering cost and improve the throughput of streaming servers significantly.
Statistical learning for effective visual information retrieval.
For effective retrieval of visual information, statistical learning plays a pivotal role. Statistical learning in such a context faces at least two major mathematical challenges: scarcity of training data, and imbalance of training classes. We present these challenges and outline our methods for addressing them: active learning, recursive subspace co-training, adaptive dimensionality reduction, class-boundary alignment, and quasi-bagging. 1
Adaptive Feature%2DSpace Conformal Transformation for Imbalanced%2DData Learning.
When the training instances of the target class are heavily outnumbered by non-target training instances, SVMs can be ineffective in determining the class boundary. To remedy this problem, we propose an adaptive conformal transformation (ACT) algorithm. ACT considers feature-space distance and the class-imbalance ratio when it performs conformal transformation on a kernel function. Experimental results on UCI and real-world datasets show ACT to be effective in improving class prediction accuracy. 1.
The anatomy of a multimodal information filter.

Image copy detection using dynamic partial function.

Image copy detection using dynamic partial function.

Framework for detecting hazardous events.

MEGA %2D the maximizing expected generalization algorithm for learning complex query concepts.
Specifying exact query concepts has become increasingly challenging to end-users. This  is because many query concepts #e.g., those for looking up a multimedia object# can be hard  to articulate, and articulation can be subjective. In this study,we propose a query-concept  learner that learns query criteria through an intelligent sampling process. Our concept learner  aims to ful#ll two primary design objectives: 1# it has to be expressive in order to model most  practical query concepts, and 2# it must learn a concept quickly and with a small number of  labeled data since online users tend to be too impatient to provide much feedback. To ful#ll  the #rst goal, we model query concepts in k-CNF, which can express almost all practical query  concepts. To ful#ll the second design goal, we propose our maximizing expected generalization  algorithm #MEGA#, which converges to target concepts quickly by its two complementary  steps: sample selection and concept re#nement. We also propose a divide-and-conquer method  that divides the concept-learning task into G subtasks to achieve speedup. We notice that a  task must be divided carefully, or search accuracy may su#er. Wethus employ a genetic-based  mining algorithm to discover good feature groupings. Through analysis and mining results,  we observe that organizing image features in a multi-resolution manner, and minimizing intragroup  feature correlation, can speed up query-concept learning substantially while maintaining  high search accuracy. Through examples, analysis, experiments, and an prototype implementation,  we show that MEGA converges to query concepts signi#cantly faster than traditional  methods.  Keywords: query concept, relevance feedback, active learning, data mining.  1 
CBSA content%2Dbased soft annotation for multimodal image retrieval using Bayes point machines.
We propose a content-based soft annotation (CBSA) procedure for providing images with semantical labels. The annotation procedure starts with labeling a small set of training images, each with one single semantical label (e.g., forest, animal, or sky). An ensemble of binary classifiers is then trained for predicting label membership for images. The trained ensemble is applied to each individual image to give the image multiple soft labels, and each label is associated with a label membership factor. To select a base binary-classifier for CBSA, we experiment with two learning methods, Support Vector Machines (SVMs) and Bayes Point Machines (BPMs), and compare their class-prediction accuracy. Our empirical study on a 116-category 25K-image set shows that the BPM-based ensemble provides better annotation quality than the SVM-based ensemble for supporting multimodal image retrievals.
Fine%2Dgrained device management in an interactive media server.
Abstract—The use of interactive media has already gained considerable popularity. Interactivity gives viewers VCR controls like slow-motion, pause, fast-forward, and instant replay. However, traditional server-based or client-based approaches for supporting interactivity either consume too much network bandwidth or require large client buffering; and hence they are economically unattractive. In this paper, we propose the architecture and design of an interactive media proxy (IMP) server that transforms noninteractive broadcast or multicast streams into interactive ones for servicing a large number of end users. For IMP to work cost-effectively, it must carefully manage its storage devices, which are needed for caching voluminous media data. In this regard, we propose a fine-grained device management strategy consisting of three complementary components: disk profiler, data placement, and IO scheduler. Through quantitative analysis and experiments, we show that these fine-grained strategies considerably improve device throughput under various workload scenarios. Index Terms—Data placement, disk profiling, interactive, IO scheduling, media server, streaming, video-on-demand.
Discovery of a perceptual distance function for measuring image similarity.

Virtual IO preemptible disk access.
Supporting preemptible disk access is essential for interactive multimedia applications that require short response time. In this study, we propose Virtual IO, an abstraction for disk IO, that transforms a nonpreemptible IO request into a preemptible one. In order to achieve its objective efficiently, Virtual IO uses disk profiling to obtain accurate and detailed knowledge about the disk. Upon implementation of Virtual IO, we show that not only does Virtual IO enable highly preemptible disk access, but it does so with little or no loss in disk throughput.
PBIR%2DMM multimodal image retrieval and annotation.


DynDex a dynamic and non%2Dmetric space indexer.
To date, almost all research work in the Content-Based Image Retrieval (CBIR) community uses Minkowski-like functions to measure similarity between images. In this paper, we first present a non-metric distance function, dynamic partial function (DPF), which works significantly better than Minkowskilike functions for measuring perceptual similarity; and we explain DPF&#039;s link to similarity theories in cognitive science. We then propose DynDex, an indexing method that deals with both the dynamic and non-metric aspects of the distance function.  DynDex employs statistical methods including distancebased classification and bagging to enable ecient indexing with DPF. In addition to its efficiency for conducting similarity searches in very high-dimensional spaces, we show that DynDex remains quite effective when features are weighted dynamically for supporting personalized searches.
An Architecture of a Web%2DBased Collaborative Image Search Engine.

MORF A Distributed Multimodal Information Filtering System.


Indexing Images in High%2DDimensional and Dynamic%2DWeighted Feature Spaces.

Clustering for Approximate Similarity Search in High%2DDimensional Spaces.
AbstractÐIn this paper, we present a clustering and indexing paradigm (called Clindex) for high-dimensional search spaces. The scheme is designed for approximate similarity searches, where one would like to find many of the data points near a target point, but where one can tolerate missing a few near points. For such searches, our scheme can find near points with high recall in very few IOs and perform significantly better than other approaches. Our scheme is based on finding clusters and, then, building a simple but efficient index for them. We analyze the trade-offs involved in clustering and building such an index structure, and present extensive experimental results. Index TermsÐApproximate search, clustering, high-dimensional index, similarity search. 1
Support vector machine active learning for image retrieval.
Relevance feedback is often a critical component when designing image databases. With these databases it is difficult to specify queries directly and explicitly. Relevance feedback interactively determinines a user’s desired output or query concept by asking the user whether certain proposed images are relevant or not. For a relevance feedback algorithm to be effective, it must grasp a user’s query concept accurately and quickly, while also only asking the user to label a small number of images. We propose the use of a support vector machine active learning algorithm for conducting effective relevance feedback for image retrieval. The algorithm selects the most informative images to query a user and quickly learns a boundary that separates the images that satisfy the user’s query concept from the rest of the dataset. Experimental results show that our algorithm achieves significantly higher search accuracy than traditional query refinement schemes after just three to four rounds of relevance feedback.
PBIR perception%2Dbased image retrieval%2Da system that can quickly capture subjective image query concepts.

SVM Binary Classifier Ensembles for Image Classification.

Spin Discriminant Analysis %28SDA%29 %2D Using a One%2DDimensional Classifier for High Dimensional Classification Problems.

Mining Image Features for Efficient Query Processing.

On Answering Queries in the Presence of Limited Access Patterns.
. In information-integration systems, source relations often  have limitations on access patterns to their data; i.e., when one must  provide values for certain attributes of a relation in order to retrieve its  tuples. In this paper we consider the following fundamental problem: can  we compute the complete answer to a query by accessing the relations  with legal patterns? The complete answer to a query is the answer that  we could compute if we could retrieve all the tuples from the relations.  We give algorithms for solving the problem for various classes of queries,  including conjunctive queries, unions of conjunctive queries, and conjunctive  queries with arithmetic comparisons. We prove the problem is  undecidable for datalog queries. If the complete answer to a query cannot  be computed, we often need to compute its maximal answer. The  second problem we study is, given two conjunctive queries on relations  with limited access patterns, how to test whether the maximal answer to...
Data Placement for Multi%2Duser Interactive DTV.
In this paper, we propose an interactive DTV design that  converts non-interactive broadcast DTV streams into interactive  ones for multiple simultaneous viewers. To enable  viewing interactivity, we show that it is critical to organize  data intelligently for improving IO resolution, reducing  disk latency and minimizing storage cost. We propose  three data placement schemes that offer different tradeoffs  between IO resolution, disk latency and storage. By employing  different schemes under different workload scenarios,  an intelligent system can minimize memory use and  hence the system cost.
Learning Image Query Concepts via Intelligent Sampling.

NNEW Nearest Neighbor Expansion By Weighting In Image Database Retrieval.

PBIR %2D Perception%2DBased Image Retrieval.

Answering queries with useful bindings.
this paper, we propose a query-planning framework to answer queries in the presence of limited access patterns. In the framework, a query and source descriptions are translated to a recursive datalog program. We then solve optimization problems in this framework, including how to decide whether accessing off-query sources is necessary, how to choose useful sources for a query, and how to test query containment. We develop algorithms to solve these problems, and thus construct an efficient program to answer a query
Query Planning with Limited Source Capabilities.
In information-integration systems, sources may have diverse and limited query capabilities. In this paper we show that because sources have restrictions on retrieving their information, sources not mentioned in a query can contribute to the query result by providing useful bindings. In some cases we can access sources repeatedly to retrieve bindings to answer a query, and query planning thus becomes considerably more challenging. We find all the obtainable answers to a query by translating the query and source descriptions to a simple recursive Datalog program, and evaluating the program on the source relations. This program often accesses sources that are not in the query. Some of these accesses are essential, as they provide bindings that let us query sources, which we could not do otherwise. However, some of these accesses can be proven not to add anything to the query&#039;s answer. We show in which cases these off-query accesses are useless, and prove that in these cases we can comput...
Maximizing QoS for interactive DTV clients.

DTV a client%2Dbased interactive DTV architecture.
Abstract In this study we present a client-based architecturethat supports time-shift Digital-TV (DTV) features (i.e., pause, replay and fast-forward) and interac-tive applications. We call this architecture ssDTVfor Personalizable Interactive Digital-TV architecture. Our study focuses on devising effective tech-niques for managing the client&#039;s data (e.g., textual, binary, and video/audio data) under the local re-source constraints. We outline the challenges that ssDTV faces and sketch data and resource man-agement policies that maximize the client&#039;s QoS adaptively based on each individual viewer&#039;s qual-ity requirement. We present a prototype that have been built and describe an array of applications that ssDTV can help realize.
MEDIC A Memory %26 Disk Cache for Multimedia Clients.
In this paper we do focus on the client side, presenting a combined memory-disk buffering algorithm that allows the client to dynamically and effectively deal with variable data rates and delays. We call this algorithm MEDIC, for MEmory-DIsk Cache. MEDIC manages the resources of the client&#039;s computer or set-top box. MEDIC carefully allocates its limited memory to competing tasks, i.e., to receiving new data from the network, to writing data to disk as memory fills up, to reading data from disk as needed, and to holding data for decoding and playback. Since data is concurrently written to the disk cache and read from the disk cache, MEDIC must also intelligently issue IOs to avoid undue conflicts.
PowerBookmarks A System for Personalizable Web Information Organization
Users of the Web are overloaded with information. This medium is “polluted ” with redundant, erroneous and low quality information. A WWW survey of 11,700 users conducted from April 10 to May 10, 1996[1] indicates that 30.31 % of the users report “finding known info ” is their problem and 27.80 % of the users report organizing collected information as their problem. An empirical study[2] on users ’ revisitation patterns to WWW pages found that 58 % of an individual’s pages are revisits. With these study results, we believe the Web users would like to build and organize a larger collection of bookmarks for future references than they can reasonably maintain now. 2. System Architecture PowerBookmarks is being developed under the auspices of a larger WebDB project to address the above issues. WebDB hypermedia database system[3] is built on top of NEC PERCIO Object-Oriented Database Manage-ment System[4]. Unlike most search engines, which fo-
2D BubbleUp Managing Parallel Disks for Media Servers.
In this study we present a scheme called two-dimensional BubbleUp (2DB) for managing parallel disks in a multimedia server. Its goal is to reduce initial latency for interactive multimedia applications, while balancing disk loads to maintain high throughput. The 2DB scheme consists of a data placement and a request scheduling policy. The data placement policy replicates frequently accessed data and places them cyclically throughout the disks. The request scheduling policy attempts to maintain free &#034;service slots&#034; in the immediate future. These slots can then be used to quickly service newly arrived requests. Through examples and simulation, we show that our scheme significantly reduces initial latency and maintains throughput comparable to that of the traditional schemes. Keywords: multimedia, data replication, initial latency, disk array. 1 Introduction  Media servers are designed to provide large numbers of presentations in the form of audios, movies or news clips. These servers need...
An Image Coding and Reconstruction Scheme for Mobile Computing.
Abstract. An asynchronous transfer mode (ATM) wireless network has bursty and high error rates. To combat the contiguous bit loss due to damaged or dropped packets, this paper presents a code packetization and image reconstruction scheme. The packetization method distributes the loss in both frequency and spatial domains to reduce the chance that adjacent DCT blocks lose the same frequency components. The image reconstruction takes into consideration the spatial characteristics represented by the frequency components. Combining these two approaches one can reconstruct the damaged images more accurately, even under very high loss rates. In addition, since the reconstruction technique is computational efficient, it conserves system resources and power consumption, which are restrictive in mobile computers. Keywords: multimedia, compression, mobile computing. 1 Introduction Many studies have proposed image and video data delivery schemes over wired channels. Since mobile computing has gained popularity in recent years, it is important also to devise robust techniques that work in a mobile environment. This paper presents image coding and reconstruction techniques that are suitable in a mobile (wireless) environment.
Cost%2DBased Media Server Design.
Conventional wisdom holds that reducing disk latency leads to higher disk utilization, maximizing disk utilization leads to higher throughput, and employing a faster disk leads to better performance. All of this is true when building a conventional file or database system. In this paper we show that these principles can be misleading when applied to a media server. To design such a server, we propose a cost-based approach that focuses on the perstream costs. We give various examples to illustrate the design process.  Keywords: multimedia, disk latency, memory utilization, per-stream cost. 1 Introduction  Maximizing throughput is a common design objective for a media server. To improve throughput, two approaches have been used: reducing disk latency (i.e., seek overhead and rotational delay) and minimizing the required memory. To reduce disk latency we can either employ efficient disk scheduling [13, 14, 15] or enact intelligent data placement policies [9, 11]. Both methods effectively ...
BubbleUp Low Latency Fast%2DScan for Media Servers.
Interactive multimedia applications require fast response time. Traditional disk scheduling schemes can incur high latencies, and caching data in memory to reduce latency is usually not feasible, especially if fast-scans need to be supported. In this study we propose a disk-based solution called BubbleUp. It significantly reduces the initial latency for new requests, as well as for fast-scan requests. The throughput of the scheme is comparable to that of traditional schemes, and it may even provide better throughput than mechanisms based on elevator disk scheduling. BubbleUp incurs a slight disk storage overhead, but we argue that through effective allocation, this cost can be minimized.
Effective Memory Use in a Media Server.
A number of techniques have been developed for maximizing disk utilization in media servers, including disk arm scheduling and data placement ones. Instead, in this paper we focus on how to efficiently utilize the available memory. We present techniques for best memory use under different disk policies, and derive precise formulas for computing memory use. We show that with proper memory use, maximizing disk utilization does not necessarily lead to optimal throughput. In addition, we study the impact of data placement policies including disk partitioning and multiple disks. Finally, our analysis shows that maximizing disk utilization and disk striping incur high system costs, and are not advisable in a media server. 
Reducing Initial Latency in Media Servers.

STeP Deductive%2DAlgorithmic Verification of Reactive and Real%2DTime Systems.

Reducing Initial Latency in a Multimedia Storage System.
A multimedia server delivers presentations (e.g., videos, movies, games), providing high bandwidth and continuous real-time delivery. In this paper we present techniques for reducing the initial latency of presentations, i.e., for reducing the time between the arrival of a request and the start of the presentation. Traditionally, initial latency has not received much attention. This is because one major application of multimedia servers is &#034;movies on demand&#034; where a delay of a few minutes before a new multi-hour movie starts is acceptable. However, latency reduction is important in interactive applications such as video games and browsing of multimedia documents. Various latency reduction schemes are proposed and analyzed, and their performance compared. We show that our techniques can significantly reduce (almost eliminate in some cases) initial latency without adversely affecting throughput. Moreover, a novel on-disk partial data replication scheme that we propose proves to be far mo...
Cost Analyses for VBR Video Servers.
In this paper we compare techniques for storage and real-time retrieval of Variable Bit Rate (VBR) video data for multiple simultaneous users. The motivation for considering VBR is that video results in inherently time varying data, and as such, with the same average bit rate, higher quality can be achieved with VBR than with Constant Bit Rate (CBR). We propose and compare the following three classes of VBR data placement and retrieval techniques: Constant Time Length (CTL) places and retrieves data in blocks corresponding to equal playback durations, Constant Data Length (CDL) places and retrieves constant-sized data blocks, and a hybrid solution uses CDL placement but retrieves a variable number of blocks in each service round. We have found that CTL data placement has much lower buffer requirements than CDL but suffers from fragmentation during video editing. We show hybrid placement to have both advantages of high efficiency and low fragmentation. We also address the issue of admis...
STeP The Stanford Temporal Prover.
We describe the Stanford Temporal Prover (STeP), a system being developed to support the computer-aided formal verification of concurrent and reactive systems based on temporal specifications. Unlike systems based on model-checking, STeP  is not restricted to finite-state systems. It combines model checking and deductive methods to allow the verification of a broad class of systems, including programs with infinite data domains, N-process programs, and N-component circuit designs, for arbitrary N . In short, STeP has been designed with the objective of combining the expressiveness of deductive methods with the simplicity of model checking. The verification process is for the most part automatic. User interaction occurs mostly at the highest, most intuitive level, primarily through a graphical proof language of verification diagrams. Efficient simplification methods, decision procedures, and invariant generation techniques are then invoked automatically to prove resulting first-order ve...
Admissions Control and Data Placement for VBR Video Servers.

Compositional Verification of Real%2DTime Systems
This paper presents a compositional proof system for the verification of real-time systems. Real-time systems are modeled as timed transition modules, which explicitly model interaction with the environment and may be combined using composition operators.  Composition rules are devised such that the correctness of a system may be determined from the correctness of its components. These proof rules are demonstrated on Fischer&#039;s mutual exclusion algorithm, for which mutual exclusion and bounded response are proven. 1 Introduction  This paper presents a compositional proof system for the verification of real-time systems. Real-time systems are a particular class of reactive systems, i.e., systems exhibiting ongoing interaction with the environment, distinguished by the observation that the correctness of a real-time system depends not only on the qualitative aspects of its behavior, but also on the quantitative timeliness of its behavior. Real-time systems are modeled as timed transition ...
Scalable Video Data Placement on Parallel Disk Arrays.

Characterization of Temporal Property Classes.
. This paper presents two novel characterizations of the classes of properties of reactive systems in terms of their expression by temporal logic. The first family of characterizations concerns the safety-progress classification, which describes a hierarchy within the set of temporal properties. Previous characterizations of this hierarchy depended critically on the use of past temporal operators. The characterization presented here identifies the future formulas that belong to each class. This characterization is shown to be complete. The second characterization concerns the safety-liveness classification, which partitions temporal properties into the classes of safety and liveness. While automata-theoretic and temporal logic characterizations of the safety class have been known for some time, a complete characterization of the liveness class by temporal logic remained open. This paper provides such a characterization.  1 Introduction  Reactive systems are systems that are expected to...
An Open and Extensible Event%2DBased Transaction Manager.

