Multiclass learning problems involve nding a de nition for an unknown function f(x) whose range is a discrete set containing k>2values (i.e., k \classes"). The de nition is acquired by studying collections of training examples of the form hx i;f(x i)i. Existing approaches to multiclass learning problems include direct application of multiclass algorithms such as the decision-tree algorithms C4.5 and CART, application of binary concept learning algorithms to learn individual binary functions for each of the k classes, and application of binary concept learning algorithms with distributed output representations. This paper compares these three approaches to a new technique in which error-correcting codes are employed as a distributed output representation. We show that these output representations improve the generalization performance of both C4.5 and backpropagation on a wide range of multiclass learning tasks. We also demonstrate that this approach is robust with respect to changes in the size of the training sample, the assignment of distributed representations to particular classes, and the application of over tting avoidance techniques such as decision-tree pruning. Finally,we show that|like the other methods|the error-correcting code technique can provide reliable class probability estimates. Taken together, these results demonstrate that error-correcting output codes provide a general-purpose method for improving the performance of inductive learning programs on multiclass problems. 1.

This study seeks to understand how network failures affect the availability of service delivery across wide area networks and to evaluate classes of techniques for improving end-to-end service availability. Using several large-scale connectivity traces, we develop a model of network unavailability that includes key parameters such as failure location and failure duration. We then use trace-based simulation to evaluate several classes of techniques for coping with network unavailability. We find that caching alone is seldom effective at insulating services from failures but that the combination of mobile extension code and prefetching can improve average unavailability by as much as an order of magnitude for classes of service whose semantics support disconnected operation. We find that routing-based techniques may provide significant improvements, but that the improvements of many individual techniques are limited because they do not address all significant categories of network failures. By combining the techniques we examine, some systems may be able to reduce average unavailability by as much as one or two orders of magnitude.

. In this paper we present a simple linear-time centroid-based document classification algorithm, that despite its simplicity and robust performance, has not been extensively studied and analyzed. Our experiments show that this centroid-based classifier consistently and substantially outperforms other algorithms such as Naive Bayesian, k-nearest-neighbors, and C4.5, on a wide range of datasets. Our analysis shows that the similarity measure used by the centroidbased scheme allows it to classify a new document based on how closely its behavior matches the behavior of the documents belonging to different classes. This matching allows it to dynamically adjust for classes with different densities and accounts for dependencies between the terms in the different classes. 1 Introduction We have seen a tremendous growth in the volume of online text documents available on the Internet, digital libraries, news sources, and company-wide intranets. It has been forecasted that these docu...

This paper presents a method for generating compact pictorial summarizations of video. We developed a novel approach for selecting still images from a video suitable for summarizing the video and for providing entry points into it. Images are laid out in a compact, visually pleasing display reminiscent of a comic book or Japanese manga. Users can explore the video by interacting with the presented summary. Links from each keyframe start video playback and/or present additional detail. Captions can be added to presentation frames to include commentary or descriptions such as the minutes of a recorded meeting. We conducted a study to compare variants of our summarization technique. The study participants judged the manga summary to be significantly better than the other two conditions with respect to their suitability for summaries and navigation, and their visual appeal.

Categorization of documents is challenging, as the number of discriminating words can be very large. We present a nearest neighbor classification scheme for text categorization in which the importance of discriminating words is learned using mutual information and weight adjustment techniques. The nearest neighbors for a particular document are then computed based on the matching words and their weights. We evaluate our scheme on both synthetic and real world documents. Our experiments with synthetic data sets show that this scheme is robust under different emulated conditions. Empirical results on real world documents demonstrate that this scheme outperforms state of the art classification algorithms such as C4.5, RIPPER, Rainbow, and PEBLS.

Web applications suffer from software and configuration faults that lower their availability. Recovering from failure is dominated by the time interval between when these faults appear and when they are detected by site operators. We introduce a set of tools that augment the ability of operators to perceive the presence of failure: an automatic anomaly detector scours HTTP access logs to find changes in user behavior that are indicative of site failures, and a visualizer helps operators rapidly detect and diagnose problems. Visualization addresses a key question of autonomic computing of how to win operators ’ confidence so that new tools will be embraced. Evaluation performed using HTTP logs from Ebates.com demonstrates that these tools can enhance the detection of failure as well as shorten detection time. Our approach is application-generic and can be applied to any Web application without the need for instrumentation. 1.

This paper presents an algorithm to segment text and picture in an image using two features based on the statistical distribution of the wavelet coefficients in high frequency bands. The algorithm breaks the image into blocks and classifies every block as background, text or picture according to the two features. The block size is variable so that the segmentation can be accurate at the boundary of two types and avoids misclassifying due to over-localized region analysis. 1 Introduction Statistical classification is an important topic in image processing. Classification, which helps to interpret an image, can also be incorporated with other image processing to improve performance. One well-known example is image compression. For training-based image compression algorithms, such as vector quantization [1], a codebook is optimally designed under the assumption that the data to be quantized is statistically consistent with the training data. Hence, different quantizers are required for ...

Numerous studies have implicated cells of the lymphoid series as agents responsible for the destruction of solid tissue homografts in vivo (1-5). This view has been substantiated by recent demonstrations that lymphoid cells (6-15) and macrophages (16, 17) procured from specifically immunized animals have a cytocidal effect on appropriate homologous "target " cells in vitro. It was previously reported that by the 6th to 7th day after grafting with homologous skiu, lymphocytes from the regional lymph nodes of mice or rats had acquired the capacity to injure foreign target renal cells in culture, and that neither isoimmune serum nor complement was required for this destructive reactivity (15). The present communication is concerned with the application of a sensitive, reproducible in vitro assay system to elucidate further quantitative aspects of the destruction of homologous target cells by sensitized lymphocytic cells. Particular attention has been paid to the influence of (a) variation in the ratio of attacking lymphoid cells to target ceils (b) the duration of contact with lymphoid cells and (c) the possible influence of the presence or absence of isoimmune

of individual cells in motor cortex. J. Neurophysiol. 77: 826–852, 1992; Soechting and Flanders 1989, 1992). These processes 1997. This study shows that the discharge of many motor cortical are often described for heuristic purposes in terms of the cells is strongly influenced by attributes of movement related to the geometry and mechanics of the arm and not only by spatial convenient but arbitrary parameter spaces (hand path, joint attributes of the hand trajectory. The activity of 619 directionally angles, joint torques, etc.) and transformations (inverse kinetuned cells was recorded from the motor cortex of two monkeys matics, inverse dynamics) of newtonian mechanics. How-during reaching movements with the use of similar hand paths but ever, it is highly unlikely that the brain controls movement two different arm orientations, in the natural parasagittal plane and by explicitly solving the newtonian laws of motion. Instead, abducted into the horizontal plane. Nearly all cells (588 of 619, psychophysical studies are revealing the physiological pa-95%) showed statistically significant changes in activity between rameters, reference frames, and transformations by which the the two arm orientations [analysis of variance (ANOVA), P õ motor system plans and implements movements (Flanders et 0.01]. A majority of cells showed a significant change in their

The events occurring in the execution of a distributed or parallel application are related by a partial, rather than a total, order. Wehave developed prototype software that collects such events during program execution and produces a graphical display consistent with the partial order. Such a display can be very helpful in understanding and debugging distributed and parallel applications. However, using only partial-order information does not allow the performance characteristics of an application to be understood. Integrating real-time information with the partial order can provide a display that is useful for understanding both functional and performance aspects of the application. An algorithm is required to adjust the collected real-time information, to ensure that real times are consistent with the partial order. Lamport's clock algorithm provides such an adjustment, but can significantly distort the real-time values. It was necessary to develop a more complex algorithm, using the same basic principles, that minimises such distortions. We have extended existing prototype software for displaying event data, so that either a purely partial-order display or a real-time displaymay be obtained. The real-time facilities can be used in multiple target environments, such as OSF/DCE, Hermes, and SR.