Showing papers by "AT&T Labs published in 2001"

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.

A quasi-orthogonal space-time block code

Abstract: It has been shown that a complex orthogonal design that provides full diversity and full transmission rate for a space-time block code is not possible for more than two antennas. Previous attempts have been concentrated in generalizing orthogonal designs which provide space-time block codes with full diversity and a high transmission rate. We design rate one codes which are quasi-orthogonal and provide partial diversity. The decoder of the proposed codes works with pairs of transmitted symbols instead of single symbols.

Convolution Kernels for Natural Language

Abstract: We describe the application of kernel methods to Natural Language Processing (NLP) problems. In many NLP tasks the objects being modeled are strings, trees, graphs or other discrete structures which require some mechanism to convert them into feature vectors. We describe kernels for various natural language structures, allowing rich, high dimensional representations of these structures. We show how a kernel over trees can be applied to parsing using the voted perceptron algorithm, and we give experimental results on the ATIS corpus of parse trees.

Graphviz: Open source graph drawing tools

Abstract: Graphviz is a heterogeneous collection of graph drawing tools containing batch layout programs (dot, neato, fdp, twopi); a platform for incremental layout (Dynagraph); customizable graph editors (dotty, Grappa); a server for including graphs in Web pages (WebDot); support for graphs as COM objects (Montage); utility programs useful in graph visualization; and libraries for attributed graphs. The software is available under an Open Source license. The article[1] provides a detailed description of the package.

Point set surfaces

Abstract: We advocate the use of point sets to represent shapes. We provide a definition of a smooth manifold surface from a set of points close to the original surface. The definition is based on local maps from differential geometry, which are approximated by the method of moving least squares (MLS). We present tools to increase or decrease the density of the points, thus, allowing an adjustment of the spacing among the points to control the fidelity of the representation. To display the point set surface, we introduce a novel point rendering technique. The idea is to evaluate the local maps according to the image resolution. This results in high quality shading effects and smooth silhouettes at interactive frame rates.

Remote State Preparation

Abstract: Quantum teleportation uses prior entanglement and forward classical communication to transmit one instance of an unknown quantum state. Remote state preparation (RSP) has the same goal, but the sender knows classically what state is to be transmitted. We show that the asymptotic classical communication cost of RSP is one bit per qubit--half that of teleportation--and even less when transmitting part of a known entangled state. We explore the tradeoff between entanglement and classical communication required for RSP, and discuss RSP capacities of general quantum channels.

Graphical models for game theory

Abstract: We introduce a compact graph-theoretic representation for multi-party game theory. Our main result is a provably correct and efficient algorithm for computing approximate Nash equilibria in one-stage games represented by trees or sparse graphs.

Voice over IP performance monitoring

Abstract: We describe a method for monitoring Voice over IP (VoIP) applications based upon a reduction of the ITU-T's E-Model to transport level, measurable quantities. In the process, 1) we identify the relevant transport level quantities, 2) we discuss the tradeoffs between placing the monitors within the VoIP gateways versus placement of the monitors within the transport path, and 3) we identify several areas where further work and consensus within the industry are required. We discover that the relevant transport level quantities are the delay, network packet loss and the decoder's de-jitter buffer packet loss. We find that an in-path monitor requires the definition of a reference de-jitter buffer implementation to estimate voice quality based upon observed transport measurements. Finally, we suggest that more studies are required, which evaluate the quality of various VoIP codecs in the presence of representative packet loss patterns.

The role of trust management in distributed systems security

Abstract: Existing authorization mechanisms fail to provide powerful and robust tools for handling security at the scale necessary for today's Internet. These mechanisms are coming under increasing strain from the development and deployment of systems that increase the programmability of the Internet. Moreover, this "increased flexibility through programmability" trend seems to be accelerating with the advent of proposals such as Active Networking and Mobile Agents. The trust-management approach to distributed-system security was developed as an answer to the inadequacy of traditional authorization mechanisms. Trust-management engines avoid the need to resolve "identities" in an authorization decision. Instead, they express privileges and restrictions in a programming language. This allows for increased flexibility and expressibility, as well as standardization of modern, scalable security mechanisms. Further advantages of the trust-management approach include proofs that requested transactions comply with local policies and system architectures that encourage developers and administrators to consider an application's security policy carefully and specify it explicitly. In this paper, we examine existing authorization mechanisms and their inadequacies. We introduce the concept of trust management, explain its basic principles, and describe some existing trust-management engines, including PoHcyMaker and KeyNote. We also report on our experience using trust-management engines in several distributed-system applications.

Approximate distance oracles

Abstract: Let G=(V,E) be an undirected weighted graph with |V|=n and |E|=m. Let k\ge 1 be an integer. We show that G=(V,E) can be preprocessed in O(kmn^{1/k}) expected time, constructing a data structure of size O(kn^{1+1/k}), such that any subsequent distance query can be answered, approximately, in O(k) time. The approximate distance returned is of stretch at most 2k-1, i.e., the quotient obtained by dividing the estimated distance by the actual distance lies between 1 and 2k-1. We show that a 1963 girth conjecture of Erd{\H{o}}s, implies that ω(n^{1+1/k}) space is needed in the worst case for any real stretch strictly smaller than 2k+1. The space requirement of our algorithm is, therefore, essentially optimal. The most impressive feature of our data structure is its constant query time, hence the name oracle. Previously, data structures that used only O(n^{1+1/k}) space had a query time of ω(n^{1/k}) and a slightly larger, non-optimal, stretch. Our algorithms are extremely simple and easy to implement efficiently. They also provide faster constructions of sparse spanners of weighted graphs, and improved tree covers and distance labelings of weighted or unweighted graphs.}

Compact routing schemes

Abstract: We describe several compact routing schemes for general weighted undirected networks. Our schemes are simple and easy to implement. The routing tables stored at the nodes of the network are all very small. The headers attached to the routed messages, including the name of the destination, are extremely short. The routing decision at each node takes constant time. Yet, the stretch of these routing schemes, i.e., the worst ratio between the cost of the path on which a packet is routed and the cost of the cheapest path from source to destination, is a small constant. Our schemes achieve a near-optimal tradeoff between the size of the routing tables used and the resulting stretch. More specifically, we obtain: A routing scheme that uses only O (n 1/2) bits of memory at each node of an n-node network that has stretch 3. The space is optimal, up to logarithmic factors, in the sense that every routing scheme with stretch n2), and every routing scheme with stretch n3/2). The headers used are only (1 + O(1)) log2> n-bits long and each routing decision takes constant time. A variant of this scheme with [log2 n] -bit headers makes routing decisions in O(log log n) time. Also, for every integer k > 2, a general handshaking based routing scheme that uses O (n1/k) bits of memory at each node that has stretch 2k - 1. A conjecture of Erdos from 1963, settled for k = 3, 5, implies that the routing tables are of near-optimal size relative to the stretch. The handshaking is similar in spirit to a DNS lookup in TCP/IP. Headers are O(log2 n) bits long and each routing decision takes constant time. Without handshaking, the stretch of the scheme increases to 4k - 5. One ingredient used to obtain the routing schemes mentioned above, may be of independent practical and theoretical interest: A shortest path routing scheme for trees of arbitrary degree and diameter that assigns each vertex of an n-node tree a (1 + O(1)) log2 n-bit label. Given the label of a source node and the label of a destination it is possible to compute, in constant time, the port number of the edge from the source that heads in the direction of the destination. The general scheme for k > 2 also uses a clustering technique introduced recently by the authors. The clusters obtained using this technique induce a sparse and low stretch tree cover of the network. This essentially reduces routing in general networks into routing problems in trees that could be solved using the above technique.

Capacity of MIMO systems with antenna selection

Abstract: We consider the capacity of multiple-input-multiple-output (MIMO) systems with reduced complexity. One link end uses all available antennas, while the other chooses the "best" L out of N antennas. As "best", we use those antennas that maximize capacity. We derive an upper bound on the capacity that can be expressed as the sum of the logarithms of ordered chi-squared variables. This bound is then evaluated analytically, and compared to results from Monte Carlo simulations. As long as L is at least as large as the number of antennas at the other link end, the achieved capacity is close to the capacity of a full-complexity system. We demonstrate, for example, that for L=3, N=8 at the receiver, and 3 antennas at the transmitter, the capacity of the reduced-complexity scheme is 20 bits/s/Hz compared to 23 bits/s/Hz of a full-complexity scheme.

Stable internet routing without global coordination

Abstract: The Border Gateway Protocol (BGP) allows an autonomous system (AS) to apply diverse local policies for selecting routes and propagating reachability information to other domains. However, the BGP permits ASs to have conflicting policies that can lead to routing instability. This paper proposes a set of guidelines for an AS to follow in setting its routing policies, without requiring coordination with other ASs. Our approach exploits the Internet's hierarchical structure and the commercial relationships between ASs to impose a partial order on the set of routes to each destination. The guidelines conform to conventional traffic-engineering practices of ISPs, and provide each AS with significant flexibility in selecting its local policies. Furthermore, the guidelines ensure route convergence even under changes in the topology and routing policies. Drawing on a formal model of BGP, we prove that following our proposed policy guidelines guarantees route convergence. We also describe how our methodology can be applied to new types of relationships between ASs, how to verify the hierarchical AS relationships, and how to realize our policy guidelines. Our approach has significant practical value since it preserves the ability of each AS to apply complex local policies without divulging its BGP configurations to others.

A Generalization of Principal Components Analysis to the Exponential Family

Abstract: Principal component analysis (PCA) is a commonly applied technique for dimensionality reduction. PCA implicitly minimizes a squared loss function, which may be inappropriate for data that is not real-valued, such as binary-valued data. This paper draws on ideas from the Exponential family, Generalized linear models, and Bregman distances, to give a generalization of PCA to loss functions that we argue are better suited to other data types. We describe algorithms for minimizing the loss functions, and give examples on simulated data.

On the constancy of internet path properties

Abstract: Many Internet protocols and operational procedures use measurements to guide future actions. This is an effective strategy if the quantities being measured exhibit a degree of constancy: that is, in some fundamental sense, they are not changing. In this paper we explore three different notions of constancy: mathematical, operational, and predictive. Using a large measurement dataset gathered from the NIMI infrastructure, we then apply these notions to three Internet path properties: loss, delay, and throughput. Our aim is to provide guidance as to when assumptions of various forms of constancy are sound, versus when they might prove misleading.

Poly-logarithmic deterministic fully-dynamic algorithms for connectivity, minimum spanning tree, 2-edge, and biconnectivity

Abstract: Deterministic fully dynamic graph algorithms are presented for connectivity, minimum spanning tree, 2-edge connectivity, and biconnectivity. Assuming that we start with no edges in a graph with n vertices, the amortized operation costs are O(log2n) for connectivity, O(log4n) for minimum spanning forest, 2-edge connectivity, and O(log5n) biconnectivity.

Deriving traffic demands for operational IP networks: methodology and experience

Abstract: Engineering a large IP backbone network without an accurate network-wide view of the traffic demands is challenging. Shifts in user behavior, changes in routing policies, and failures of network elements can result in significant (and sudden) fluctuations in load. In this paper, we present a model of traffic demands to support traffic engineering and performance debugging of large Internet Service Provider networks. By defining a traffic demand as a volume of load originating from an ingress link and destined to a set of egress links, we can capture and predict how routing affects the traffic traveling between domains. To infer the traffic demands, we propose a measurement methodology that combines flow-level measurements collected at all ingress links with reachability information about all egress links. We discuss how to cope with situations where practical considerations limit the amount and quality of the necessary data. Specifically, we show how to infer interdomain traffic demands using measurements collected at a smaller number of edge links-the peering links connecting the neighboring providers. We report on our experiences in deriving the traffic demands in the AT&T IP BAckbone, by collecting, validating, and joining very large and diverse sets of usage, configuration, and routing data over extended periods of time. The paper concludes with a preliminary analysis of the observed dynamics of the traffic demands and a discussion of the practical implications for traffic engineering.

Issues for routing in the optical layer

Abstract: Optical layer control planes based on MPLS and other Internet protocols hold great promise because of their proven scalability, ability to support rapid provisioning, and auto discovery and self-inventory capabilities and are under intense study in various standards bodies. To date however little attention has been paid to aspects of the optical layer which differ from those found in data networking. We study three such aspects which impact routing: network elements which are reconfigurable, but in constrained ways; transmission impairments which may make some routes unusable; and diversity. We conclude that if emerging optical technology is to be maximally exploited, heterogeneous technologies with dissimilar routing constraints are likely. Four alternative architectures for dealing with this eventuality are identified and some trade-offs between centralizing or distributing some aspects of routing are discussed.

Priced Oblivious Transfer: How to Sell Digital Goods

Abstract: We consider the question of protecting the privacy of customers buying digital goods. More specifically, our goal is to allow a buyer to purchase digital goods from a vendor without letting the vendor learn what, and to the extent possible also when and how much, it is buying. We propose solutions which allow the buyer, after making an initial deposit, to engage in an unlimited number of priced oblivious-transfer protocols, satisfying the following requirements: As long as the buyer's balance contains sufficient funds, it will successfully retrieve the selected item and its balance will be debited by the item's price. However, the buyer should be unable to retrieve an item whose cost exceeds its remaining balance. The vendor should learn nothing except what must inevitably be learned, namely, the amount of interaction and the initial deposit amount (which imply upper bounds on the quantity and total price of all information obtained by the buyer). In particular, the vendor should be unable to learn what the buyer's current balance is or when it actually runs out of its funds. The technical tools we develop, in the process of solving this problem, seem to be of independent interest. In particular, we present the first one-round (two-pass) protocol for oblivious transfer that does not rely on the random oracle model (a very similar protocol was independently proposed by Naor and Pinkas [21]). This protocol is a special case of a more general "conditional disclosure" methodology, which extends a previous approach from [11] and adapts it to the 2-party setting.

A random graph model for power law graphs

Abstract: We propose a random graph model which is a special case of sparserandom graphs with given degree sequences which satisfy a power law. This model involves only a small number of paramo eters, called logsize and log-log growth rate. These parameters capture some universal characteristics of massive graphs. From these parameters, various properties of the graph can be derived. For example, for certai n ranges of the parameters, we wi II compute the expected distribution of the sizes of the connected components which almost surely occur with high probability. We illustrate the consistency of our model with the behavior of some massive graphs derived from data in telecommunications. We also discuss the threshold function, the giant component, and the evolution of random graphs in this model.

WordsEye: an automatic text-to-scene conversion system

Abstract: Natural language is an easy and effective medium for describing visual ideas and mental images. Thus, we foresee the emergence of language-based 3D scene generation systems to let ordinary users quickly create 3D scenes without having to learn special software, acquire artistic skills, or even touch a desktop window-oriented interface. WordsEye is such a system for automatically converting text into representative 3D scenes. WordsEye relies on a large database of 3D models and poses to depict entities and actions. Every 3D model can have associated shape displacements, spatial tags, and functional properties to be used in the depiction process. We describe the linguistic analysis and depiction techniques used by WordsEye along with some general strategies by which more abstract concepts are made depictable.

Applying the harmonic plus noise model in concatenative speech synthesis

Abstract: This paper describes the application of the harmonic plus noise model (HNM) for concatenative text-to-speech (TTS) synthesis. In the context of HNM, speech signals are represented as a time-varying harmonic component plus a modulated noise component. The decomposition of a speech signal into these two components allows for more natural-sounding modifications of the signal (e.g., by using different and better adapted schemes to modify each component). The parametric representation of speech using HNM provides a straightforward way of smoothing discontinuities of acoustic units around concatenation points. Formal listening tests have shown that HNM provides high-quality speech synthesis while outperforming other models for synthesis (e.g., TD-PSOLA) in intelligibility, naturalness, and pleasantness.

Finding interesting associations without support pruning

Abstract: Association-rule mining has heretofore relied on the condition of high support to do its work efficiently. In particular, the well-known a priori algorithm is only effective when the only rules of interest are relationships that occur very frequently. However, there are a number of applications, such as data mining, identification of similar Web documents, clustering, and collaborative filtering, where the rules of interest have comparatively few instances in the data. In these cases, we must look for highly correlated items, or possibly even causal relationships between infrequent items. We develop a family of algorithms for solving this problem, employing a combination of random sampling and hashing techniques. We provide analysis of the algorithms developed and conduct experiments on real and synthetic data to obtain a comparative performance analysis.

Trajectory sampling for direct traffic observation

Abstract: Traffic measurement is a critical component for the control and engineering of communication networks. We argue that traffic measurement should make it possible to obtain the spatial flow of traffic through the domain, i.e., the paths followed by packets between any ingress and egress point of the domain. Most resource allocation and capacity planning tasks can benefit from such information. Also, traffic measurements should be obtained without a routing model and without knowledge of network state. This allows the traffic measurement process to he resilient to network failures and state uncertainty. We propose a method that allows the direct inference of traffic flows through a domain by observing the trajectories of a subset of all packets traversing the network. The key advantages of the method are that (1) it does not rely on routing state; (2) its implementation cost is small; and (3) the measurement reporting traffic is modest and can be controlled precisely. The key idea of the method is to sample packets based on a hash function computed over the packet content. Using the same hash function will yield the same sample set of packets in the entire domain, and enables us to reconstruct packet trajectories.

On the use and performance of content distribution networks

Abstract: Content distribution networks (CDNs) are a mechanism to deliver content to end users on behalf of origin Web sites. Content distribution offloads work from origin servers by serving some or all of the contents of Web pages. We found an order of magnitude increase in the number and percentage of popular origin sites using CDNs between November 1999 and December 2000.In this paper we discuss how CDNs are commonly used on the Web and define a methodology to study how well they perform. A performance study was conducted over a period of months on a set of CDN companies employing the techniques of DNS redirection and URL rewriting to balance load among their servers. Some CDNs generally provide better results than others when we examine results from a set of clients. The performance of one CDN company clearly improved between the two testing periods in our study due to a dramatic increase in the number of distinct servers employed in its network. More generally, the results indicate that use of a DNS lookup in the critical path of a resource retrieval does not generally result in better server choices being made relative to client response time in either average or worst case situations.

An indexed model of recursive types for foundational proof-carrying code

Abstract: The proofs of "traditional" proof carrying code (PCC) are type-specialized in the sense that they require axioms about a specific type system. In contrast, the proofs of foundational PCC explicitly define all required types and explicitly prove all the required properties of those types assuming only a fixed foundation of mathematics such as higher-order logic. Foundational PCC is both more flexible and more secure than type-specialized PCC.For foundational PCC we need semantic models of type systems on von Neumann machines. Previous models have been either too weak (lacking general recursive types and first-class function-pointers), too complex (requiring machine-checkable proofs of large bodies of computability theory), or not obviously applicable to von Neumann machines. Our new model is strong, simple, and works either in λ-calculus or on Pentiums.

An Adaptive Version of the Boost by Majority Algorithm

Abstract: We propose a new boosting algorithm. This boosting algorithm is an adaptive version of the boost by majority algorithm and combines bounded goals of the boost by majority algorithm with the adaptivity of AdaBoost. The method used for making boost-by-majority adaptive is to consider the limit in which each of the boosting iterations makes an infinitesimally small contribution to the process as a whole. This limit can be modeled using the differential equations that govern Brownian motion. The new boosting algorithm, named BrownBoost, is based on finding solutions to these differential equations. The paper describes two methods for finding approximate solutions to the differential equations. The first is a method that results in a provably polynomial time algorithm. The second method, based on the Newton-Raphson minimization procedure, is much more efficient in practice but is not known to be polynomial.

On computing correlated aggregates over continual data streams

Abstract: In many applications from telephone fraud detection to network management, data arrives in a stream, and there is a need to maintain a variety of statistical summary information about a large number of customers in an online fashion. At present, such applications maintain basic aggregates such as running extrema values (MIN, MAX), averages, standard deviations, etc., that can be computed over data streams with limited space in a straightforward way. However, many applications require knowledge of more complex aggregates relating different attributes, so-called correlated aggregates. As an example, one might be interested in computing the percentage of international phone calls that are longer than the average duration of a domestic phone call. Exact computation of this aggregate requires multiple passes over the data stream, which is infeasible.We propose single-pass techniques for approximate computation of correlated aggregates over both landmark and sliding window views of a data stream of tuples, using a very limited amount of space. We consider both the case where the independent aggregate (average duration in the example above) is an extrema value and the case where it is an average value, with any standard aggregate as the dependent aggregate; these can be used as building blocks for more sophisticated aggregates. We present an extensive experimental study based on some real and a wide variety of synthetic data sets to demonstrate the accuracy of our techniques. We show that this effectiveness is explained by the fact that our techniques exploit monotonicity and convergence properties of aggregates over data streams.