AT&T Labs

About: AT&T Labs is a based out in . It is known for research contribution in the topics: Network packet & The Internet. The organization has 1879 authors who have published 5595 publications receiving 483151 citations.

On unbiased sampling for unstructured peer-to-peer networks

Abstract: This paper presents a detailed examination of how the dynamic and heterogeneous nature of real-world peer-to-peer systems can introduce bias into the selection of representative samples of peer properties (e.g., degree, link bandwidth, number of files shared). We propose the Metropolized Random Walk with Backtracking (MRWB) as a viable and promising technique for collecting nearly unbiased samples and conduct an extensive simulation study to demonstrate that our technique works well for a wide variety of commonly-encountered peer-to-peer network conditions. We have implemented the MRWB algorithm for selecting peer addresses uniformly at random into a tool called ion-sampler. Using the Gnutella network, we empirically show that ion-sampler. yields more accurate samples than tools that rely on commonly-used sampling techniques and results in dramatic improvements in efficiency and scalability compared to performing a full crawl.

Comments on the performance of measurement-based admission control algorithms

Abstract: Relaxed real time services that do not provide guaranteed loss rates or delay bounds are of considerable interest in the Internet, since these services can achieve higher utilization than hard real time services while still providing adequate service to adaptive real-time applications. Achieving this higher level of utilization depends on an admission control algorithm that does not rely on worst-case bounds to guide its admission decisions. Measurement-based admission control is one such approach, and several measurement-based admission control algorithms have been proposed in the literature. In this paper, we use simulations to compare the performance of several of these algorithms. We find that all of them achieve nearly the same utilization for a given packet loss rate, and that none of them are capable of accurately meeting loss targets.

Random Domino Tilings and the Arctic Circle Theorem

Abstract: In this article we study domino tilings of a family of finite regions called Aztec diamonds. Every such tiling determines a partition of the Aztec diamond into five sub-regions; in the four outer sub-regions, every tile lines up with nearby tiles, while in the fifth, central sub-region, differently-oriented tiles co-exist side by side. We show that when n is sufficiently large, the shape of the central sub-region becomes arbitrarily close to a perfect circle of radius n/sqrt(2) for all but a negligible proportion of the tilings. Our proof uses techniques from the theory of interacting particle systems. In particular, we prove and make use of a classification of the stationary behaviors of a totally asymmetric one-dimensional exclusion process in discrete time.

Reducing Multiclass to Binary: A Unifying Approach for Margin Classifiers

Abstract: We present a unifying framework for studying the solution of multiclass categorization problems by reducing them to multiple binary problems that are then solved using a margin-based binary learning algorithm. The proposed framework unifies some of the most popular approaches in which each class is compared against all others, or in which all pairs of classes are compared to each other, or in which output codes with error-correcting properties are used. We propose a general method for combining the classifiers generated on the binary problems, and we prove a general empirical multiclass loss bound given the empirical loss of the individual binary learning algorithms. The scheme and the corresponding bounds apply to many popular classification learning algorithms including support-vector machines, AdaBoost, regression, logistic regression and decision-tree algorithms. We also give a multiclass generalization error analysis for general output codes with AdaBoost as the binary learner. Experimental results with SVM and AdaBoost show that our scheme provides a viable alternative to the most commonly used multiclass algorithms.

Optimizing SVMs for complex call classification

Abstract: Large margin classifiers such as support vector machines (SVM) or Adaboost are obvious choices for natural language document or call routing. However, how to combine several binary classifiers to optimize the whole routing process and how this process scales when it involves many different decisions (or classes) is a complex problem that has only received partial answers. We propose a global optimization process based on an optimal channel communication model that allows a combination of possibly heterogeneous binary classifiers. As in Markov modeling, computational feasibility is achieved through simplifications and independence assumptions that are easy to interpret. Using this approach, we have managed to decrease the call-type classification error rate for AT&T's How May I Help You (HMIHY/sup (sm)/) natural dialog system by 50 %.