Catálogo de publicaciones - libros

Software failure and software reliability are strongly related concepts. Introducing a model that would perform successful failure prediction could provide the means for achieving higher software reliability and quality. In this context, we have employed artificial neural networks and genetic algorithms to investigate whether software failure can be accurately modeled and forecasted based on empirical data of real systems.

This paper suggests several estimation guidelines for the choice of a suitable machine learning technique for software development effort estimation. Initially, the paper presents a review of relevant published studies, pointing out pros and cons of specific machine learning methods. The techniques considered are Association Rules, Classification and Regression Trees, Bayesian Belief Networks, Neural Networks and Clustering, and they are compared in terms of accuracy, comprehensibility, applicability, causality and sensitivity. Finally the study proposes guidelines for choosing the appropriate technique, based on the size of the training data and the desirable features of the extracted estimation model.

The number of software defects found in software applications today costs users and companies billions of dollars annually. In general, these defects occur due to an inadequate software development process that does not give the necessary importance to testing. Another contributor to these costs is the lack of adequate automated tools that can find “bugs” that would not otherwise be verified by experts. This paper looks at the combinatorial characteristics of the problem of testing — tools essentially search among all test cases for those that are promising (find existing bugs in the application) — and the effect that abstractions inspired by nature, such as genetic algorithms and swarm intelligence, may have in the construction of more “intelligent” testing tools. The paper argues that these abstractions may be used to construct automated tools that are more powerful, less biased, and able to incorporate expert knowledge while maintaining the ability to discover new, never-thought-of software defects.

One of the most common problems in computer vision and image processing applications is the localization of object boundaries in a video frame and its tracking in the next frames. In this paper, a fully automatic method for fast tracking of video objects in a video sequence using affine invariant normalization is proposed. Initially, the detection of a video object is achieved using a GVF snake. Next, a vector of the affine parameters of each contour of the extracted video object in two successive frames is computed using affine-invariant normalization. Under the hypothesis that these contours are similar, the affine transformation between the two contours is computed in a very fast way. Using this transformation to predict the position of the contour in the next frame allows initialization of the GVF snake very close to the real position. Applying this technique to the following frames, a very fast tracking technique is achieved. Moreover, this technique can be applied on sequences with very fast moving objects where traditional trackers usually fail. Results on synthetic sequences are presented which illustrate the theoretical developments.

We propose an approach to knowledge acquisition, which uses multimedia ontologies for fused extraction of semantics from multiple modalities, and feeds back the extracted information, aiming to evolve knowledge representation. This paper presents the basic components of the proposed approach and discusses the open research issues focusing on the fused information extraction that will enable the development of scalable and precise knowledge acquisition technology.

This paper relates semantics as it is used in linguistics and natural language processing to the operational requirements of image retrieval systems. This is done in the context of a model of exploratory search and image annotation or indexing. The paper concludes this operational context requires the use of a restricted form of semantics compared with the usual one from linguistics or natural language processing, focussing on words rather sentences.

Most of the successful commercial applications in language processing (text and/or speech) dispense of any explicit concern on semantics, with the usual motivations stemming from the computational high costs required by dealing with semantics in case of large volumes of data. With recent advances in corpus linguistics and statistical-based methods in NLP, revealing useful semantic features of linguistic data is becoming cheaper and cheaper and the accuracy of this process is steadily improving. Lately, there seems to be a growing acceptance of the idea that multilingual lexical ontologies might be the key towards aligning different views on the semantic atomic units to be used in characterizing the general meaning of various and multilingual documents. Depending on the granularity at which semantic distinctions are necessary, the accuracy of the basic semantic processing (such as word sense disambiguation) can be very high with relatively low complexity computing. The paper substantiates this statement by presenting a statistical/based system for word alignment (WA) and word sense disambiguation (WSD) in parallel corpora.

Space-time tubes, a feature that can be used for analysis of motion based on the observed moving points in a scene is introduced. Information provided by sensors is used to detect moving points and based on their connectivity, tubes enable a structured approach towards identifying moving objects and high level events. It is shown that using tubes in conjunction with domain knowledge can overcome errors caused by the inaccuracy or inadequacy of the original motion information. The detected high level events can then be mapped to small natural language descriptions of object motion in the scene.

In this paper, we consider the automatic identification of video shots that are relevant to a given semantic concept from large video databases. We apply a method of representing semantic concepts as class models on a set of parallel Self-Organizing Maps trained with multimodal low-level features. The presented experiments were conducted using a set of 170 hours of video containing recorded television news programs.

An art installation was on display in the Centre Pompidou National Museum of Modern Art in Paris, where visitors could contribute with their own personal objects, adding keyword descriptions and quantified semantic features such as or . The data was projected in real-time onto a Self-Organizing Map (SOM) which was shown in the gallery. In this paper we analyze the same data by extracting visual features from the images and organize the image collection with multiple SOMs. We show how this mapping facilitates the emergence of semantic associations between visual, textual and metadata modalities by studying the distributions of the different feature vectors on the SOMs.