Search

In this paper, we reduce the energy overheads of continuous mobile sensing, specifically for the case of context-aware applications that are interested in collective context or events , i.e., events expressed as a set of complex predicates over sensor data from multiple smartphones. We propose a cloud-based query management and optimization framework, called CloQue , that can support thousands of such concurrent queries, executing over a large number of individual smartphones. Our central insight is that the context of different individuals & groups often have significant correlation, and that this correlation can be learned through standard association rule mining on historical data. CloQue’s exploits such correlation to reduce energy overheads via two key innovations: (i) dynamically reordering the order of predicate processing to preferentially select predicates with not just lower sensing cost and higher selectivity, but that maximally reduce the uncertainty about other context predicates; and (ii) intelligently propagating the query evaluation results to dynamically update the confidence values of other correlated context predicates. We present techniques for probabilistic processing of context queries (to save significant energy at the cost of a query fidelity loss) and for query partitioning (to scale CloQue to a large number of users while meeting latency bounds). An evaluation, using real cellphone traces from two different datasets, shows significant energy savings (between 30% and 50% compared with traditional short-circuit systems) with little loss in accuracy (5% at most). In addition, we utilize parallel evaluation to reduce overall latency. The experiments show our approaches save up to 70% latency.

Plant phenotyping systems strive to maintain high categorization accuracy when expanding their scopes to larger environments. In this paper, we discuss problems associated with expanding the plant categorization scope. These problems are particularly complicated due to the increase in the number of species and the inter-species similarity. In our approach, we modify previously trained Convolutional Neural Networks (CNNs) and integrate domain-specific knowledge in the fine-tuning process of these models to maintain high accuracy while expanding the scope. This process is the key idea behind our CNN-based expanding approach resulting in plant-expert models. Experiments described in this paper compare the accuracy of an expanded phenotyping system using different plant-related datasets during the training of the CNN categorization models. Although it takes much longer to train these models, our approach achieves better performance compared to models trained without the integration of domain-specific knowledge, especially when the number of species increases significantly.

Given a set of query points within an image coordinate system, constellation queries identify the matching points in a database of known points within a standard coordinate system. Constellation queries are an integral part of orientation determination systems used in spacecrafts to orient and navigate themselves. The query points are bright spots in an image captured by a camera on the spacecraft and the database contains known celestial objects in a celestial coordinate system. This paper studies six existing constellation query processing strategies (Angle, Interior Angle, Spherical Triangle, Planar Triangle, Pyramid, Composite Pyramid) using a unified algorithmic framework and presents experimental evaluation of the six strategies. We find that the Pyramid strategy in its simplified form has the best accuracy to runtime ratio given simulated images with false positives, false negatives, and Gaussian noise.

Automatic categorization of plant species in natural images is an important computer vision problem with numerous applications in agriculture and botany. The problem is particularly challenging due to the large number of plant species, the inter-species similarity, the large scale variations in natural images, and the lack of annotated data. In this paper, we present a guided multi-scale approach that segments the regions of interest (containing a plant) from a complex background of the natural image and systematically extracts scale-representative patches based on those regions. These multi-scale patches are used to train state-of-the-art Convolutional Neural Network (CNN) models that analyze a given plant image and determine its species. Focusing specifically on the identification of plant species in natural images, we show that the proposed approach is a very effective way of making deep learning models more robust to scale variations. We perform a comprehensive experimental evaluation of our proposed method over several CNN models. Our best result on the Inception-ResNet-v2 model achieves a top-1 classification accuracy of 89.21% for 100 plant species which represents a 5.4% increase over using random cropping to generate training data.

Despite the availability of dozens of plant identification mobile applications, identifying plants from a natural image remains a challenging problem - most of the existing applications do not address the complexity of natural images, the large number of plant species, and the multi-scale nature of natural images. In this technical demonstration, we present the WTPlant system for identifying plants in natural images. WTPlant is based on deep learning approaches. Specifically, it uses stacked Convolutional Neural Networks for image segmentation, a novel preprocessing stage for multi-scale analyses, and deep convolutional networks to extract the most discriminative features. WTPlant employs different classification architectures for plants and flowers, thus enabling plant identification throughout all the seasons. The user interface also shows, in an interactive way, the most representative areas in the image that are used to predict each plant species. The first version of WTPlant is trained to classify 100 different plant species present in the campus of the University of Hawai'i at Manoa. First experiments support the hypothesis that an initial segmentation process helps guide the extraction of representative samples and, consequently, enables Convolutional Neural Networks to better recognize objects of different scales in natural images. Future versions aim to extend the recognizable species to cover the land-based flora of the Hawaiian Islands.

A key operation in data cleaning and integration is the use of string similarity join (SSJ) algorithms to identify and remove duplicates or similar records within data sets. With the advent of big data, a natural question is how to parallelize SSJ algorithms. There is a large body of existing work on SSJ algorithms and parallelizing each one of them may not be the most feasible solution. In this paper, we propose a parallelization framework for string similarity joins that utilizes existing SSJ algorithms. Our framework partitions the data using a variety of partitioning strategies and then executes the SSJ algorithms on the partitions in parallel. Some of the partitioning strategies that we investigate trade accuracy for speed. We implemented and validated our framework on several SSJ algorithms and data sets. Our experiments show that our framework results in significant speedup with little loss in accuracy.

In database systems, the cost of data storage and retrieval are important components of the total cost and response time of the system. A popular mechanism to reduce the storage footprint is by compressing the data residing in tables and indexes. Compressing indexes efficiently, while maintaining response time requirements, is known to be challenging. This is especially true when designing for a workload spectrum covering both data warehousing and transaction processing environments. DB2 Linux, UNIX, Windows (LUW) recently introduced index compression for use in both environments. This uses techniques that are able to compress index data efficiently while incurring virtually no performance penalty for query processing. On the contrary, for certain operations, the performance is actually better. In this paper, we detail the design of index compression in DB2 LUW and discuss the challenges that were encountered in meeting the design goals. We also demonstrate its effectiveness by showing performance results on typical customer scenarios.

One of the key advantages of cloud computing is the elasticity in which computing resources such as virtual machines can be increased or decreased. Current state-of-the-art shared-nothing parallel SQL processing systems, on the other hand, are often designed and optimized for a fixed number of database nodes. To take advantage of the elasticity afforded by cloud computing, cloud-based SQL processing systems need the ability to repartition the data easily when the number of database nodes is scaled up or down. In this paper, we investigate the problem of supporting elastic partitioning of data in cloud-based parallel SQL processing systems. We propose several algorithms and associated data organization techniques that minimizes the re-partitioning of tuples and the movement of data between nodes. Our experimental evaluation demonstrates the effectiveness of the proposed methods.

Many emerging context-aware mobile applications involve the execution of continuous queries over sensor data streams generated by a variety of on-board sensors on multiple personal mobile devices (aka smartphones). To reduce the energy-overheads of such large-scale, continuous mobile sensing and query processing, this paper introduces CQP, a collaborative query processing framework that exploits the overlap (in both the sensor sources and the query predicates) across multiple smartphones. The framework automatically identifies the shareable parts of multiple executing queries, and then reduces the overheads of repetitive execution and data transmissions, by having a set of `leader' mobile nodes execute and disseminate these shareable partial results. To further reduce energy, CQP utilizes lower-energy short-range wireless links (such as Bluetooth) to disseminate such results directly among proximate smartphones. We describe algorithms to support our server-assisted distributed query sharing and optimization strategy. Simulation experiments indicate that this approach can result in 60% reduction in the energy overhead of continuous query processing; when `leader' selection is dynamically rotated to equitably share the burden, we observe an increase of up to 65% in operational lifetime.

We make the case for cloud-enabled mobile sensing services that support an emerging application class, one which infers near-real time collective context using sensor data obtained continuously from a large set of consumer mobile devices. We present the high-level architecture and functional requirements for such a mobile sensing service, and argue that such a service can significantly improve the scalability and energy-efficiency of large-scale mobile sensing by coordinating the sensing & processing tasks across multiple devices. We then focus specifically on the problem of energy-efficiency and provide early exemplars of how optimizing query execution jointly over multiple phones can lead to substantial energy savings.

To make semantic search a reality, we need to be able to efficiently publish large data sets containing rich semantic structure. We have tools for translating relational and semi-structured data into RDF, but such translation tools do not have the goal of adding or providing the kind of semantics necessary to achieve the goals of the Semantic Web and semantic search over the Web. In this chapter, we present LinQuer, a tool for creating semantic links within a data source and between data sources. We focus on link discovery over structured (relational) data since many Semantic Web sources are the result of publishing relational data as RDF and since relational engines provide the scalability and flexibility we need for large scale link discovery. The LinQuer framework is based on the declarative specification of linkage requirements by a user. We present algorithms for translating these requirements to queries that can run over relational data sources, potentially using semantic information (such as a class hierarchy or a more general ontology) to enhance the recall of the link discovery. We show that this framework is flexible enough to permit linking real data, including dirty data (which is commonly found on the Web) and data with a variety of semantic connections.

An increasing number of applications operate on data obtained from the Web. These applications typically maintain local copies of the web data to avoid network latency in data accesses. As the data on the Web evolves, it is critical that the local copy be kept up-to-date. Data freshness is one of the most important data quality issues, and has been extensively studied for various applications including web crawling. However, web crawling is focused on obtaining as many raw web pages as possible. Our applications, on the other hand, are interested in specific content from specific data sources. Knowing the content or the semantics of the data enables us to differentiate data items based on their importance and volatility, which are key factors that impact the design of the data synchronization strategy. In this work, we formulate the concept of content freshness, and present a novel approach that maintains content freshness with least amount of web communication. Specifically, we assume data is accessible through a general keyword search interface, and we form keyword queries based on their selectivity, as well their contribution to content freshness of the local copy. Experiments show the effectiveness of our approach compared with several naive methods for keeping data fresh.

Discovering links between different data items in a single data source or across different data sources is a challenging problem faced by many information systems today. In particular, the recent Linking Open Data (LOD) community project has highlighted the paramount importance of establishing semantic links among web data sources. Currently, LOD sources provide billions of RDF triples, but only millions of links between data sources. Many of these data sources are published using tools that operate over relational data stored in a standard RDBMS. In this paper, we present a framework for discovery of semantic links from relational data. Our framework is based on declarative specification of linkage requirements by a user. We illustrate the use of our framework using several link discovery algorithms on a real world scenario. Our framework allows data publishers to easily find and publish high-quality links to other data sources, and therefore could significantly enhance the value of the data in the next generation of web.

Remote monitoring of patients' biometric data streams offers the possibility to physicians to extend and improve their services to chronically ill patients who are away from medical institutions. This emerging technology is a promising way to address important aspects of the cost issues that most health care systems are experiencing. In order to fulfill its potential, several challenges need to be overcome. First, the data collected needs to be filtered and annotated intelligently to help physicians cope with and navigate the large amount of patient sensor data received as a result of large scale remote health monitoring deployments. Secondly, efficient stream persistence and query mechanisms for these data need to be designed to satisfy health care regulations and help physicians track patient health histories accurately and efficiently. In this paper, we concentrate on the second challenge. We leverage emerging hybrid relational-XML database management systems to design a storage sub-system for remote health monitoring. We evaluate this approach by performing series of performance tests to assess the ability of the proposed system to handle the huge amount of biometric data streams requiring persistence.

Electronic commerce is emerging as a major application area for database systems. A large number of e-commerce stores provide electronic product catalogs that allow customers to search products of interest and store owners to manage various product information. Due to the constant schema evolution and the sparsity of e-commerce data, most commercial e-commerce systems use the so-called vertical schema for data storage. However, query processing for data stored using vertical schema is extremely inefficient because current RDBMSs, especially its cost-based query optimizer, are specifically designed to deal with traditional horizontal schemas. In this paper, we show that e-catalog management can be naturally supported in IBM's system RX, the first DBMS that truly supports both XML and relational data in their native forms. By leveraging on system RX's hybrid nature, we present a novel solution for storing, managing, and querying e-catalog data. In addition to traditional queries, we show that our solution supports semantic queries as well. Our solution does not require a separate query optimization layer, because query optimization is handled within the hybrid DBMS engine itself

The World Wide Web is growing and changing at an astonishing rate. For the information in the web to be useful, web information systems such as search engines have to keep up with the growth and change of the web. In this paper we study how web documents change. In particular, we study two important characteristics of web document change that are directly related to keeping web information systems upto-date: the degree of the change and the clusteredness of the change. We analyze the evolution of web documents with respect to these two measures and discuss the implications for web information systems update.