The concepts of conditional mutual information (CMI) and normalized CMI (NCMI) are introduced to measure the concentration and separation performance of a classification deep neural network (DNN) in the output probability distribution space of the DNN, where CMI and the ratio between CMI and NCMI represent the intraclass concentration and interclass separation of the DNN, respectively. By using NCMI to evaluate popular DNNs pretrained over CIFAR-100 and ImageNet in the literature, it is shown that their validation accuracies are more or less inversely proportional to their NCMI values. Based on this observation, the standard deep learning (DL) framework is further modified to minimize the standard cross entropy (CE) function subject to an NCMI constraint, yielding CMI constrained DL (CMIC-DL). A novel alternating learning algorithm is proposed to solve such a constrained optimization problem. Extensive experimental results show that DNNs trained within CMIC-DL outperform the state-of-the-art models trained within the standard DL and other loss functions in the literature in terms of both accuracy and robustness against adversarial attacks. In addition, visualizing the evolution of the learning process through the lens of CMI and NCMI is also advocated.

In this paper, we introduce the notions of conditional mutual information (CMI) and normalized conditional mutual information (NCMI) for classification deep neural networks (DNNs). In particular, CMI and the ratio between CMI and NCMI quantify the intra-class concentration and inter-class separation of a DNN in its output probability distribution space, respectively. Utilizing NCMI to assess widely recognized DNNs pre-trained on ImageNet reveals a notable inverse relationship between their validation accuracies and NCMI values on the ImageNet validation dataset. Building upon this insight, the conventional deep learning (DL) framework is modified by minimizing the standard cross-entropy function while imposing an NCMI constraint. This refinement results in a novel approach known as CMI-constrained deep learning (CMIC-DL). Comprehensive experimental findings demonstrate that DNNs trained using CMIC-DL outperform state-of-the-art models trained within standard DL and other loss functions in the literature. In addition, some semantic meaning of CMI is also discovered.

Abstract — As the academic world moves away from physical journals and proceedings towards online document repositories, the ability to efficiently locate work of interest among the torrent of newly-generated papers will become increasingly important. To aid in this endeavor, we designed SmartSeer, a system that allows users to register personalized continuous queries over the CiteSeer database of technical documents. Users are then alerted whenever papers that match their queries are put online. SmartSeer has two main design requirements. First, to allow effective information retrieval, it should support rich continuous queries (as opposed to simple keyword searches). Second, to make effective use of donated infrastructure, it should be capable of running on a loosely maintained group of unreliable machines spread across multiple organizations (as opposed to assuming a reliable and tightly coupled distributed system). Existing work on distributed continuous query systems fails at least one of these requirements. Our design for SmartSeer is based on Distributed Hash Tables (DHTs), and thereby leverages previous work on DHT-based query systems. A prototype of SmartSeer has been implemented and evaluated on Planetlab. Though we evaluate our design only for the SmartSeer application, we believe it also provides useful insights into other distributed and rich continuous query systems (web alerts, news alerts etc). I.

Major search engines currently use the history of a user's actions (e.g., queries, clicks) to personalize search results. In this paper, we present a new personalized service, query-specific web recommendations (QSRs), that retroactively answers queries from a user's history as new results arise. The QSR system addresses two important subproblems with applications beyond the system itself: (1) Automatic identification of queries in a user's history that represent standing interests and unfulfilled needs. (2) Effective detection of interesting new results to these queries. We develop a variety of heuristics and algorithms to address these problems, and evaluate them through a study of Google history users. Our results strongly motivate the need for automatic detection of standing interests from a user's history, and identifies the algorithms that are most useful in doing so. Our results also identify the algorithms, some which are counter-intuitive, that are most useful in identifying interesting new results for past queries, allowing us to achieve very high precision over our data set.

Holistic twig join algorithms represent the state of the art for evaluating path expressions in XML queries. Using inverted indexes on XML elements, holistic twig joins move a set of index cursors in a coordinated way to quickly find structural matches. Because each cursor move can trigger I/O, the performance of a holistic twig join is largely determined by how many cursor moves it makes, yet, surprisingly, existing join algorithms have not been optimized along these lines. In this paper, we describe TwigOptimal, a new holistic twig join algorithm with optimal cursor movement. We sketch the proof of TwigOptimal's optimality, and describe how TwigOptimal can use information in the return clause of XQuery to boost its performance. Finally, experimental results are presented, showing TwigOptimal's superiority over existing holistic twig join algorithms.

Large-scale competitive P2P networks are threatened by the non-cooperation problem, where peers do not forward queries to potential competitors. Non-cooperation will be a growing problem in such applications as pay-per-transaction file-sharing, P2P auctions, and P2P service discovery networks, where peers are in competition with each other to provide services. Here, the authors showed how non-cooperation causes unacceptable degradation in quality of results, and present an economic protocol to address this problem. This protocol, called the RTR protocol, is based on the buying and selling of the right-to-respond (RTR) to each query in the network. Through simulations it is shown how the RTR protocol not only overcomes non-cooperation by providing proper incentives to peers, but also results in a network that is even more effective and efficient through intelligent, incentive-compatible routing of messages

Structural joins are a fundamental operation in XML query processing and a large body of work has focused on index-based algorithms for executing them. In this paper, we describe how two well-known index features -- path indices and ancestor information -- can be combined in a novel way to replace one or more of the physical index cursors in a structural join with virtual cursors. The position of a virtual cursor is derived from the path and ancestor information of a physical cursor. Implementation results are provided to show that, by eliminating index I/O, virtual cursors can improve the performance of structural joins by an order of magnitude or more.

Current search techniques over unstructured peer-to-peer networks rely on intelligent forwarding-based techniques to propagate queries to other peers in the network. Forwarding techniques are attractive because they typically require little state and offer robustness to peer failures; however they have inherent performance drawbacks due to the overhead of forwarding and lack of central control. We study GUESS, a nonforwarding search mechanism, as a viable alternative to currently popular forwarding-based mechanisms. We show how non-forwarding mechanisms can be over an order of magnitude more efficient than forwarding mechanisms; however, they must be deployed with care, as a naive implementation can result in highly suboptimal performance, and make them susceptible to hotspots and misbehaving peers.

In addition to indexes and materialized views, horizontal and vertical partitioning are important aspects of physical design in a relational database system that significantly impact performance. Horizontal partitioning also provides manageability; database administrators often require indexes and their underlying tables partitioned identically so as to make common operations such as backup/restore easier. While partitioning is important, incorporating partitioning makes the problem of automating physical design much harder since: (a) The choices of partitioning can strongly interact with choices of indexes and materialized views. (b) A large new space of physical design alternatives must be considered. (c) Manageability requirements impose a new constraint on the problem. In this paper, we present novel techniques for designing a scalable solution to this integrated physical design problem that takes both performance and manageability into account. We have implemented our techniques and evaluated it on Microsoft SQL Server. Our experiments highlight: (a) the importance of taking an integrated approach to automated physical design and (b) the scalability of our techniques. 1.

Peer-to-peer systems enable access to data spread over an extremely large number of machines. Most P2P systems support only simple lookup queries. However, many new applications, such as P2P photo sharing and massively multi-player games, would benefit greatly from support for multidimensional range queries. We show how such queries may be supported in a P2P system by adapting traditional spatial-database technologies with novel P2P routing networks and load-balancing algorithms. We show how to adapt two popular spatial-database solutions - kd-trees and space-filling curves - and experimentally compare their effectiveness.

Emerging economic P2P applications share the common need for an efficient, secure payment mechanism. In this paper, we present PPay, a micropayment system that exploits unique characteristics of P2P systems to maximize efficiency while maintaining security properties. We show how the basic PPay protocol far outperforms existing micropayment schemes, while guaranteeing that all coin fraud is detectable, traceable and unprofitable. We also present and analyze several extensions to PPay that further improve efficiency.

A super-peer is a node in a peer-to-peer network that operates both as a server to a set of clients, and as an equal in a network of super-peers. Super-peer networks strike a balance between the efficiency of centralized search, and the autonomy, load balancing and robustness to attacks provided by distributed search. Furthermore, they take advantage of the heterogeneity of capabilities (e.g., bandwidth, processing power) across peers, which recent studies have shown to be enormous. Hence, new and old P2P systems like KaZaA and Gnutella are adopting super-peers in their design. Despite their growing popularity, the behavior of super-peer networks is not well understood. For example, what are the potential drawbacks of super-peer networks? How can super-peers be made more reliable? How many clients should a super-peer take on to maximize efficiency? we examine super-peer networks in detail, gaming an understanding of their fundamental characteristics and performance tradeoffs. We also present practical guidelines and a general procedure for the design of an efficient super-peer network.

Peer-to-peer systems have emerged as a popular way to share huge volumes of data. The usability of these systems depends on effective techniques to find and retrieve data; however current techniques used in existing P2P systems are often very inefficient. We present three techniques for efficient search in P2P systems. We present the design of these techniques, and then evaluate them using a combination of analysis and experiments over Gnutella, the largest open P2P system in operation. We show that while our techniques maintain the same quality of results as currently used techniques, they use up to 5 times fewer resources. In addition, we designed our techniques to be simple, so that they can be easily incorporated into existing systems for immediate impact.

We identify crucial design issues in building a distributed inverted index for a large collection of web pages. We introduce a novel pipelining technique for structuring the core index-building system that substantially reduces the index construction time. We also propose a storage scheme for creating and managing inverted files using an embedded database system. We propose and compare different strategies for addressing various issues relevant to distributed index construction.

We identify crucial design issues in building a distributed inverted index for a large collection of Web pages. We introduce a novel pipelining technique for structuring the core index-building system that substantially reduces the index construction time. We also propose a storage scheme for creating and managing inverted files using an embedded database system. We suggest and compare different strategies for collecting global statistics from distributed inverted indexes. Finally, we present performance results from experiments on a testbed distributed Web indexing system that we have implemented.