The diversity of intrinsic qualities of multimedia entities tends to impede their effective retrieval. In a Self-Learning Search Engine architecture, the subtle nuances of human perceptions and deep knowledge are taught and captured through unsupervised reinforcement learning, where the degree of reinforcement may be suitably calibrated. Such architectural paradigm enables indexes to evolve naturally while accommodating the dynamic changes of user interests. It operates by continuously constructing indexes over time, while injecting progressive improvement in search performance. For search operations to be effective, convergence of index learning is of crucial importance to ensure efficiency and robustness. In this paper, we develop a Self-Learning Search Engine architecture based on reinforcement learning using a Markov Decision Process framework. The balance between exploration and exploitation is achieved through evolutionary exploration Strategies. The evolutionary index learning behavior is then studied and formulated using stochastic analysis. Experimental results are presented which corroborate the steady convergence of the index evolution mechanism.

In the search and retrieval of multimedia objects, it is impractical to either manually or automatically extract the contents for indexing since most of the multimedia contents are not machine extractable, while manual extraction tends to be highly laborious and time-consuming. However, by systematically capturing and analyzing the feedback patterns of human users, vital information concerning the multimedia contents can be harvested for effective indexing and subsequent search. By learning from the human judgment and mental evaluation of users, effective search indices can be gradually developed and built up, and subsequently be exploited to find the most relevant multimedia objects. To avoid hovering around a local maximum, we apply the -greedy method to systematically explore the search space. Through such methodic exploration, we show that the proposed approach is able to guarantee that the most relevant objects can always be discovered, even though initially it may have been overlooked or not regarded as relevant. The search behavior of the present approach is quantitatively analyzed, and closed-form expressions are obtained for the performance of two variants of the -greedy algorithm, namely EGSE-A and EGSE-B. Simulations and experiments on real data set have been performed which show good agreement with the theoretical findings. The present method is able to leverage exploration in an effective way to significantly raise the performance of multimedia information search, and enables the certain discovery of relevant objects which may be otherwise undiscoverable.

Musical signals have rich temporal information not only at the physical level but at the emotion level. The listeners may wish to find music excerpts that have similar sequence patterns of musical emotions with given excerpts. Most state-of-the-art systems for emotion-based music retrieval concentrate on static analysis of musical emotions, and ignore dynamic analysis and modeling of musical emotions overtime. This paper presents a novel approach to perform music retrieval based on time-varying musical emotion dynamics. A three-dimensional musical emotion model-Resonance-Arousal-Valence (RAV)-is used, and emotions of a piece of music are represented by musical emotion dynamics in a time series. A multiple dynamic textures (MDT) model is proposed to model music and emotion dynamics overtime, and expectation maximization (EM) algorithm along with Kalman filtering and smoothing is used to estimate model parameters. Two smoothing methods-Rauch-Tung-Striebel (RTS) and minimum-variance smoothing (MVS)-to robust model are investigated and compared to find an optimal solution to enhance prediction. To find similar sequence patterns of musical emotions, subsequence dynamic time warping (DTW) for emotion dynamics matching is presented. Experimental results demonstrate the benefits of MDT to predict time-varying musical emotions, and our proposed method for music retrieval based on emotion dynamics outperforms retrieval methods based on acoustic features.

We present several interrelated technical and empirical contributions to the problem of emotion-based music recommendation and show how they can be applied in a possible usage scenario. The contributions are (1) a new three-dimensional resonance-arousal-valence model for the representation of emotion expressed in music, together with methods for automatically classifying a piece of music in terms of this model, using robust regression methods applied to musical/acoustic features; (2) methods for predicting a listener’s emotional state on the assumption that the emotional state has been determined entirely by a sequence of pieces of music recently listened to, using conditional random fields and taking into account the decay of emotion intensity over time; and (3) a method for selecting a ranked list of pieces of music that match a particular emotional state, using a minimization iteration method. A series of experiments yield information about the validity of our operationalizations of these contributions. Throughout the article, we refer to an illustrative usage scenario in which all of these contributions can be exploited, where it is assumed that (1) a listener’s emotional state is being determined entirely by the music that he or she has been listening to and (2) the listener wants to hear additional music that matches his or her current emotional state. The contributions are intended to be useful in a variety of other scenarios as well.

Recent findings show that online reviews, blogs, and discussion forums on chronic diseases and drugs are becoming important supporting resources for patients. Extracting information from these substantial bodies of texts is useful and challenging. We developed a generative probabilistic aspect mining model (PAMM) for identifying the aspects/topics relating to class labels or categorical meta-information of a corpus. Unlike many other unsupervised approaches or supervised approaches, PAMM has a unique feature in that it focuses on finding aspects relating to one class only rather than finding aspects for all classes simultaneously in each execution. This reduces the chance of having aspects formed from mixing concepts of different classes; hence the identified aspects are easier to be interpreted by people. The aspects found also have the property that they are class distinguishing: They can be used to distinguish a class from other classes. An efficient EM-algorithm is developed for parameter estimation. Experimental results on reviews of four different drugs show that PAMM is able to find better aspects than other common approaches, when measured with mean pointwise mutual information and classification accuracy. In addition, the derived aspects were also assessed by humans based on different specified perspectives, and PAMM was found to be rated highest.

Effective sharing of diverse social media is often inhibited by limitations in their search and discovery mechanisms, which are particularly restrictive for media that do not lend themselves to automatic processing or indexing. Here, we present the structure and mechanism of an adaptive search engine which is designed to overcome such limitations. The basic framework of the adaptive search engine is to capture human judgment in the course of normal usage from user queries in order to develop semantic indexes which link search terms to media objects semantics. This approach is particularly effective for the retrieval of multimedia objects, such as images, sounds, and videos, where a direct analysis of the object features does not allow them to be linked to search terms, for example, nontextual/icon-based search, deep semantic search, or when search terms are unknown at the time the media repository is built. An adaptive search architecture is presented to enable the index to evolve with respect to user feedback, while a randomized query-processing technique guarantees avoiding local minima and allows the meaningful indexing of new media objects and new terms. The present adaptive search engine allows for the efficient community creation and updating of social media indexes, which is able to instill and propagate deep knowledge into social media concerning the advanced search and usage of media resources. Experiments with various relevance distribution settings have shown efficient convergence of such indexes, which enable intelligent search and sharing of social media resources that are otherwise hard to discover.

This tutorial will be based on the recently published book, High-Performance Parallel Database Processing and Grid Databases (John Wiley & Sons, 2008). The sizes of databases have seen exponential growth in the past and such growth is expected to accelerate in the future, with the steady drop in storage cost accompanied by a rapid increase in storage capacity. To effectively manage such volumes of data, it is necessary to allocate multiple resources to it, very often massively so. The processing of databases of such astronomical proportions requires an understanding of how high performance systems and parallelism work. Besides the massive volume of data in the database to be processed, some data has been distributed across the globe in a Grid environment. This important new book provides readers with a fundamental understanding of parallelism in data-intensive applications, and demonstrates how to develop faster capabilities to support them. It features not only the algorithms for database operations, but also quantitative analytical models, so that performance can be analyzed and evaluated more effectively.

Adaptive search engines (ASE), used in the retrieval of multimedia objects adapt their behavior depending on the user feedback in order to eventually converge to the optimal answer. The adaptive architecture has been shown to improve the performance in case of multimedia objects retrieval, when pre-indexing techniques are costly or can be applied only partially. The continuous user feedbacks on the lists of returned objects are used to filter out irrelevant objects and promote the relevant ones. This work propose an original dealer/opponent game model for ASE. The system/user interactive process which takes place in ASE can be modeled as a discovery game between a dealer, the user community which holds a secret consisting in the optimal answer to a query, and an opponent, i.e. the system, which tries to discover the secret by submitting tentative solutions on which it receives the user/dealer feedback. It is shown how the complexity of the game can be related to known games. An evolutionary approach to solve the ASE game is also presented. Experimental results shows convergence to the optimal solution with acceptable performance for real domain size. The proposed schema is quite general and can fit other adaptive search architectures which appear in e-business and e-commerce applications.

As the number of web images is increasing at a rapid rate, searching them semantically presents a significant challenge. Many raw images are constantly uploaded with little meaningful direct annotations of semantic content, limiting their search and discovery. In this paper, we present a semantic annotation technique based on the use of image parametric dimensions and metadata. Using decision trees and rule induction, we develop a rule-based approach to formulate explicit annotations for images fully automatically, so that by the use of our method, semantic query such as " sunset by the sea in autumn in New York" can be answered and indexed purely by machine. Our system is evaluated quantitatively using more than 100,000 web images. Experimental results indicate that this approach is able to deliver highly competent performance, attaining good recall and precision rates of sometimes over 80%. This approach enables a new degree of semantic richness to be automatically associated with images which previously can only be performed manually.

Performance analysis of agent network topologies helps multi-agent system developers to understand the impact of topology on system efficiency and effectiveness. Appropriate topology analysis enables the adoption of suitable frameworks for the specific multi-agent systems. In this paper, we propose a novel hybrid topology for distributed multi-agent systems, and compare the performance of this topology with two other common agent network topologies within the new multi-agent framework, Agent-based Open Connectivity for DSS (AOCD). Three major aspects are studied for estimating topology performance, which include (i) transmission time for a set of requests; (ii) waiting time for processing requests; and (iii) memory consumption for storing agent information.

Image indexing and retrieval using a concept-based approach involves extraction, modelling and indexing of image content information. Computer vision offers a variety of techniques for searching images in large collections. We propose a method that enables components of an image to be categorised on the basis of their relative importance in combination with filtered representations. Our method concentrates on matching subparts of images, defined in a variety of ways, in order to find particular objects. These ideas are illustrated with a variety of examples. We focus on Concept-based Image Indexing and Retrieval (CIIR), using a fuzzy expert systems, density measure, supporting factors and other attributes of image components to identify and retrieve images accurately and efficiently.

This paper focuses on implicit concept-based image indexing and retrieval (ICIIR), and the development of an improved method for the indexing and retrieval of images. The method involves the development of techniques to enable components of an image to be categorised on the basis of their relative importance with the image. Thus the storage of images involves an implicit, rather than an explicit, indexing scheme. Retrieval of images will then be effected by application of an algorithm based on the categorisation, which will allow relevant images to be identified and retrieved accurately and efficiently.

Abstract Keyword search of multimedia collections lacks precision and automatic parsing of unrestricted natural language annotations lacks accuracy. We propose a structure for natural language descriptions of the semantic content of visual materials that requires descriptions to be (modified) keywords, phrases, or simple sentences, with components that are grammatical relations common to many languages. This structure makes it easy to implement a collection's descriptions as a relational database, enabling efficient search via the application of well‐developed database‐indexing methods. Description components may be elements from external resources (thesaurus, ontology, database, or knowledge base). This provides a rich superstructure for the meaningful retrieval of images by their semantic contents.

The occurrence of transaction failures is often unavoidable in most database processing environments and such failures can have a particularly detrimental effect on long-duration transactions. In this paper, the impact of failures on the performance of long-duration database transactions is studied, and results for the transaction response time and failure overhead are presented for general failure patterns, which include Poisson failures as a special case. It is found that the failure overhead can be excessive for long-duration transactions and the transaction response time can increase sharply with increased failure rates. The benefits of transaction nesting to arrest performance deterioration is considered and results for quantifying the extent of restructuring improvement are provided. A procedure based on branch and bound techniques for determining the optimal restructuring strategy to minimize total transaction response time is also presented. It is shown that restructuring optimization can bring about significant performance advantages.

A high-performance parallel system for processing databases is presented, which adopts a distributed memory architecture and has been successfully implemented on a transputer platform. In addition to developing and implementing a variety of rules and schemes for parallelizing database queries, general analytic models for distributed memory database processing have been formulated, which have been successfully validated against measurements. Experimental data also indicate that the system is able to attain near linear speedup for the join operation. We also demonstrate that a linear speedup is unattainable for distributed memory database systems, and an upper bound governing the maximum transaction rate is derived. The present system is also reconfigurable and adopts a novel processor allocation scheme based on equalizing the elapsed time among different cooperating processors, which is able to readjust the processing resources assigned to particular database tasks in accordance with changing demands on the system.

As a result of insertions and deletions, a file tends to be cluttered with deleted records which are physically present. These unwanted records cause fragmentation within the file and give rise to additional access overhead because they have to be skipped over during processing. A connection between the dynamic fragmentation characteristics and the pattern of record insertions and deletions over time is presented, and performance degradation is studied in terms of the number of record accesses per reference. Deterioration characteristics are obtained for nonhomogeneous Poisson insertion and general deletion processes. For constant insertion rate, it is found that the deterioration over time is asymptotically linear, with the rate of decline governed by the record deletion rate. An expression for the optimum compaction interval is also given for files subject to a constant insertion rate and an exponentially distributed record lifetime.

The number of records in a file system is often recognised as a key determinant of efficiency. For example, the performance of sequential processing is O(N(t)) and that of tree search is O(logN(t)), where N(t) is the number of records in the file at time t. In this paper, the growth behaviour of files is studied in terms of quite general record insertion and deletion characteristics, and the performance evolution of some of the common systems is analysed. The growth data of an actual system are compared with the model results and reasonable agreement is observed.

Computer disc accesses necessitate the mechanical movement of a read/write head over a sequence of storage locations. The pattern of such random movement has a significant bearing on access times and its estimation is of primary importance to the meaningful prediction of performance. The procedure presented, which is based on a random walk description, allows such estimates to be derived quickly, incorporating any prior knowledge and partial information on data characteristics. Although only an approximation, it is able to produce good agreement with published measurements.

Data currency and consistency for the single primary update and moving primary update strategies for replicated distributed databases are analysed. Data currency and consistency are expressed in terms of probabilities and their evaluation is based on a quantitative analysis of the race condition to which the system is subject; the race condition is dependent on the intensity of site transaction traffic and the speed of update propagation. Both homogeneous and heterogeneous systems are considered. A common requirement of resilient systems is dynamic site re-configuration which may, for example, be necessitated by an earlier site crash. Here algorithms for configuration optimisation are presented for both strategies which are simple and practical to implement; such optimisation could dramatically increase data currency and consistency, and improvements of 30% to 100+% are not untypical. In addition, the present evaluation also allows the quantitative justification of a number of intuitively known facts relating to data currency and consistency.

The execution of long-running batch programs imposes severe reliability constraints on a computing system since the occurrence of a failure during its execution is more likely and that once occurred, a failure would destroy all the processing perfonned thus far. This paper studies the execution delay and machine resources consumed in supporting the running of large batch programs in a computing environment interrupted by failures. The effect of checkpoints and their optimal insertion are also considered. The results are applicable to arbitrary law of failure.

The fragmented storage map is represented by a Markov model which incorporates correlation between locations not necessarily adjacent to one another. Formulae for the average access distance for contiguous and non-contiguous allocations are given. For large storage requirements, the performance penalty of the former can be substantially higher than that of the latter: they are shown to be O(kn) and O(n)respectively, where n is the storage requirement and k>1. The Markov model is also able to achieve close agreement with published measurements.

A model for disc locality referencing is presented which, unlike previous models, allows device efficiency to be assessed in real time. Two types of localizations are distinguished: inter-cylinder localization and intra-cylinder localization. The present study allows these factors to be quantitatively analysed. The results of the model are compared with empirical measurements, and close agreement is obtained.

The logical relationships among records in a stored database induce a corresponding structure among disc locations. Database access is manifested as cylinder address sequences conforming to certain statistical patterns. Markov chains have been previously employed to represent empirical seek patterns, and although they provide a useful first approximation, they break down when the effects of detailed implementation features need to be studied. The present approach permits these features to be naturally incorporated. It can be adapted to study systems with arbitrary reference patterns and provides a versatile and economic means for the practical performance assessment of disc databases.

Fragmentation of storage is a common phenomenon in both main storage and secondary storage. Fragmentation in secondary storage not only jeopardizes the allocation of space but also, since secondary storage access time–unlike that in main storage–is typically nonuniform, a decrease in efficiency arising from additional head movement may also result. A fragmented storage exhibits a checkerboard like pattern with free and occupied space alternating one another. Such alternating storage configuration is analyzed using alternating renewal processes. Two main types of storage processing are distinguished: contiguous storage allocation and noncontiguous storage allocation. The latter allows a request to be scattered over different locations while the former requires it to be allocated in a single continuous area. It is found that the reduction in operating efficiency due to fragmentation is quite substantial for both types of processing. The deterioration is especially marked in the former and is strongly affected by 1) the request size and 2) the storage utilization. Expressions for the generating functions of the performance penalties are derived. The results of the model are compared with published measurements and satisfactory agreement is obtained.

A simple model based on the theory of runs is used to study the degradation in disc file system performance introduced by storage fragmentation. The results of the model are compared with measured values and reasonable agreement is observed. Performance degradation in terms of extended access distance and I/O response time is found to be substantial even under moderate fragmentation.

A class of procedures which select for allocation the first feasible hole having size not exceeding the actual request size plus a tolerance factor is proposed. This class of procedures, which includes the best-fit and first-fit procedures as special cases, results from an improvement of Campbell's optimal-fit procedure and exhibits optimality with respect to the combined criteria of search delay and best-fitness. Simulation experiments indicate that these procedures compete very well with the best-fit and first-fit procedures and can outperform them in certain aspects.