Causality-informed machine learning has been proposed as an avenue for achieving many of the goals of modern machine learning, from ensuring generalization under domain shifts to attaining fairness, robustness, and interpretability. A key component of causal machine learning is the inference of causal structures from observational data; in practice, this data may be incompletely observed. Prior work has demonstrated that adversarial perturbations of completely observed training data may be used to force the learning of inaccurate structural causal models (SCMs). However, when the data can be audited for correctness (e.g., it is cryptographically signed by its source), this adversarial mechanism is invalidated. This work introduces a novel attack methodology wherein the adversary deceptively omits a portion of the true training data to bias the learned causal structures in a desired manner (under strong signed sample input validation, this behavior seems to be the only strategy available to the adversary). Under this model, theoretically sound attack mechanisms are derived for the case of arbitrary SCMs, and a sample-efficient learning-based heuristic is given. Experimental validation of these approaches on real and synthetic datasets, across a range of SCMs from the family of additive noise models (linear Gaussian, linear non-Gaussian, and non-linear Gaussian), demonstrates the effectiveness of adversarial missingness attacks at deceiving popular causal structure learning algorithms.

Causality-informed machine learning has been proposed as an avenue for achieving many of the goals of modern machine learning, from ensuring generalization under domain shifts to attaining fairness, robustness, and interpretability. A key component of causal machine learning is the inference of causal structures from observational data; in practice, this data may be incompletely observed. Prior work has demonstrated that adversarial perturbations of completely observed training data may be used to force the learning of inaccurate causal structural models (SCMs). However, when the data can be audited for correctness (e.g., it is cryptographically signed by its source), this adversarial mechanism is invalidated. This work introduces a novel attack methodology wherein the adversary deceptively omits a portion of the true training data to bias the learned causal structures in a desired manner (under strong signed sample input validation, this behavior seems to be the only strategy available to the adversary). Under this model, theoretically sound attack mechanisms are derived for the case of arbitrary SCMs, and a sample-efficient learning-based heuristic is given. Experimental validation of these approaches on real and synthetic data sets demonstrates the effectiveness of adversarial missingness attacks at deceiving popular causal structure learning algorithms.

This work considers the canonical polyadic decomposition (CPD) of tensors using proximally regularized sketched alternating least squares algorithms. First, it establishes a sublinear rate of convergence for proximally regularized sketched CPD algorithms under two natural conditions that are known to be satisfied by many popular forms of sketching. Second, it demonstrates that the iterative nature of CPD algorithms can be exploited algorithmically to choose more performant sketching rates. This is accomplished by introducing CPD-MWU, a proximally-regularized sketched alternating least squares algorithm that adaptively selects the sketching rate at each iteration. On both synthetic and real data we observe that for noisy tensors CPD-MWU produces decompositions of comparable accuracy to the standard CPD decomposition in less time, often half the time; for ill-conditioned tensors, given the same time budget, CPD-MWU produces decompositions with an order-of-magnitude lower relative error. For a representative real-world dataset CPD-MWU produces residual errors on average 20% lower than CPRAND-MIX and 44% lower than SPALS, two recent sketched CPD algorithms.

With the increased deployment of IoT and edge devices into commercial and user networks, these devices have become a new threat vector for malware authors. It is imperative to protect these devices as they become more prevalent in commercial and personal networks. However, due to their limited computational power and storage space, especially in the case of battery-powered devices, it is infeasible to deploy state-of-the-art malware detectors onto these systems. In this work, we propose using and extracting features from Markov matrices constructed from opcode traces as a low cost feature for unobfuscated and obfuscated malware detection. We empirically show that our approach maintains a high detection rate while consuming less power than similar work.

There has been an increased interest in the application of convolutional neural networks for image based malware classification, but the susceptibility of neural networks to adversarial examples allows malicious actors to evade classifiers. Adversarial examples are usually generated by adding small perturbations to the input that are unrecognizable to humans, but the same approach is not effective with malware. In general, these perturbations cause changes in the byte sequences that change the initial functionality or result in un-executable binaries. We present a generative model for executable adversarial malware examples using obfuscation that achieves a high misclassification rate, up to 100% and 98% in white-box and black-box settings respectively, and demonstrates transferability. We further evaluate the effectiveness of the proposed method by reporting insignificant change in the evasion rate of our adversarial examples against popular defense strategies.

Canonical Polyadic Decomposition (CPD) is a powerful technique for uncovering multilinear relationships in tensors. Current research in scalable CPD has focused on designing efficient decomposition algorithms for large sparse tensors that arise in machine learning and data mining applications. This work addresses the complementary need for efficient decomposition algorithms for large dense tensors that arise in signal processing applications. Such tensors are often highly skewed, with one mode (e.g., time) orders of magnitude larger than the others. We present an algorithm appropriate for MapReduce settings that uses both regularization and sketching to efficiently operate on such tensors. We have open-sourced an Apache Spark implementation of the algorithm and evaluate it on synthetic and real datasets to characterize the trade-offs in runtime and accuracy when using different types and combinations of regularization and sketching. We observe that a combination of random entry sketching plus Tikhonov regularization works best independent of the type or level of noise in the tensor. Similarly, we find that random entry sketching plus proximal regularization works best for ill-conditioned tensors. Further experiments demonstrate that the runtime scales sublinearly with the tensor size and highly sublinearly with the tensor rank. The use of regularization and sketching results in runtimes that are factors of 42-112× faster than those of the previous state-of-the-art MapReduce CPD implementation for large dense, skewed tensors, while having a negligible impact on the accuracy of the decompositions.

Time series classification using deep neural networks, such as convolutional neural networks (CNN), operate on the spectral decomposition of the time series computed using a preprocessing step. This step can include a large number of hyperparameters, such as window length, filter widths, and filter shapes, each with a range of possible values that must be chosen using time and data intensive cross-validation procedures. We propose the wavelet deconvolution (WD) layer as an efficient alternative to this preprocessing step that eliminates a significant number of hyperparameters. The WD layer uses wavelet functions with adjustable scale parameters to learn the spectral decomposition directly from the signal. Using backpropagation, we show the scale parameters can be optimized with gradient descent. Furthermore, the WD layer adds interpretability to the learned time series classifier by exploiting the properties of the wavelet transform. In our experiments, we show that the WD layer can automatically extract the frequency content used to generate a dataset. The WD layer combined with a CNN applied to the phone recognition task on the TIMIT database achieves a phone error rate of 18.1\%, a relative improvement of 4\% over the baseline CNN. Experiments on a dataset where engineered features are not available showed WD+CNN is the best performing method. Our results show that the WD layer can improve neural network based time series classifiers both in accuracy and interpretability by learning directly from the input signal.

Parallel Factor Analysis (PARAFAC) is used in many scientific disciplines to decompose multimodal datasets ('tensors') into principal factors to uncover multilinear relationships in the data. Today's popular implementations of PARAFAC are single-server solutions that do not scale well to big datasets. This paper presents the design, implementation, and testing of a Big Data-enabled Parallel PARAFAC algorithm written in Java and run via Hadoop. To optimize performance across diverse datasets, three computational modes have been implemented: (i) dense, (ii) sparse, and (iii) hybrid. The input tensor is first divided into slices for parallel processing, then a one-pass test is performed on each slice to determine its size and sparsity, and finally, the best mode for each slice is chosen to minimize the overall runtime. For tensors with non-uniform data density distributions, we demonstrate that a per slice selection of the computational mode minimizes the runtime compared to using the same mode for all slices. In one representative example, the runtime with auto-selected mode was 3% and 25% faster than pure hybrid and sparse mode executions, respectively, resulting in a faster execution of PARAFAC on Big Data tensors than previously possible.

The lack of computational support has significantly slowed down automatic understanding of endangered languages. In this paper, we take Nyushu (simplified Chinese: 女书; literally: “women’s writing”) as a case study to present the first computational approach that combines Computer Vision and Natural Language Processing techniques to deeply understand an endangered language. We developed an end-to-end system to read a scanned hand-written Nyushu article, segment it into characters, link them to standard characters, and then translate the article into Mandarin Chinese. We propose several novel methods to address the new challenges introduced by noisy input and low resources, including Nyushu-specific feature selection for character segmentation and linking, and character linking lattice based Machine Translation. The end-to-end system performance indicates that the system is a promising approach and can serve as a standard benchmark.

This study considers the problem of describing and predicting cleft formation during the early stages of branching morphogenesis in mouse submandibular salivary glands (SMG) under the influence of varied concentrations of epidermal growth factors (EGF). Given a time-lapse video of a growing SMG, first we build a descriptive model that captures the underlying biological process and quantifies the ground truth. Tissue-scale (global) and morphological features related to regions of interest (local features) are used to characterize the biological ground truth. Second, we devise a predictive growth model that simulates EGF-modulated branching morphogenesis using a dynamic graph algorithm, which is driven by biological parameters such as EGF concentration, mitosis rate, and cleft progression rate. Given the initial configuration of the SMG, the evolution of the dynamic graph predicts the cleft formation, while maintaining the local structural characteristics of the SMG. We determined that higher EGF concentrations cause the formation of higher number of buds and comparatively shallow cleft depths. Third, we compared the prediction accuracy of our model to the Glazier-Graner-Hogeweg (GGH) model, an on-lattice Monte-Carlo simulation model, under a specific energy function parameter set that allows new rounds of de novo cleft formation. The results demonstrate that the dynamic graph model yields comparable simulations of gland growth to that of the GGH model with a significantly lower computational complexity. Fourth, we enhanced this model to predict the SMG morphology for an EGF concentration without the assistance of a ground truth time-lapse biological video data; this is a substantial benefit of our model over other similar models that are guided and terminated by information regarding the final SMG morphology. Hence, our model is suitable for testing the impact of different biological parameters involved with the process of branching morphogenesis in silico, while reducing the requirement of in vivo experiments.

Key extraction via measuring a physical quantity is a class of information theoretic key exchange protocols that rely on the physical characteristics of the communication channel, to enable the computation of a shared key by two parties that share no prior secret information. The key is supposed to be information theoretically hidden to an eavesdropper. Despite the recent surge of research activity in the area, concrete claims about the security of the protocols typically rely on channel abstractions that are not fully experimentally substantiated. In this paper, we propose a novel methodology for the experimental security analysis of these protocols. The crux of our methodology is a falsifiable channel abstraction that is accompanied by an efficient experimental approximation algorithm of the conditional min-entropy available to the parties given the view of the eavesdropper. We focus on the signal strength between two wirelessly communicating transceivers as the measured quantity, and we use an experimental setup to compute the conditional min-entropy of the channel given the view of the attacker which we find to be linearly increasing. Armed with this understanding of the channel, we showcase the methodology by providing a general protocol for key extraction in this setting that is shown to be secure for a concrete parameter selection. In this way, we provide a comprehensively analyzed wireless key extraction protocol that is demonstrably secure against passive adversaries assuming our falsifiable channel abstraction. Our use of hidden Markov models as the channel model and a dynamic programming approach to approximate conditional min-entropy might be of independent interest, while other possible instantiations of our methodology can be feasible and may be motivated by this paper.

Boliang Zhang, Hongzhao Huang, Xiaoman Pan, Sujian Li, Chin-Yew Lin, Heng Ji, Kevin Knight, Zhen Wen, Yizhou Sun, Jiawei Han, Bulent Yener. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

Cleft formation during submandibular salivary gland branching morphogenesis is the critical step initiating the growth and development of the complex adult organ. Previous experimental studies indicated requirements for several epithelial cellular processes, such as proliferation, migration, cell-cell adhesion, cell-extracellular matrix (matrix) adhesion, and cellular contraction in cleft formation; however, the relative contribution of each of these processes is not fully understood since it is not possible to experimentally manipulate each factor independently. We present here a comprehensive analysis of several cellular parameters regulating cleft progression during branching morphogenesis in the epithelial tissue of an early embryonic salivary gland at a local scale using an on lattice Monte-Carlo simulation model, the Glazier-Graner-Hogeweg model. We utilized measurements from time-lapse images of mouse submandibular gland organ explants to construct a temporally and spatially relevant cell-based 2D model. Our model simulates the effect of cellular proliferation, actomyosin contractility, cell-cell and cell-matrix adhesions on cleft progression, and it was used to test specific hypotheses regarding the function of these parameters in branching morphogenesis. We use innovative features capturing several aspects of cleft morphology and quantitatively analyze clefts formed during functional modification of the cellular parameters. Our simulations predict that a low epithelial mitosis rate and moderate level of actomyosin contractility in the cleft cells promote cleft progression. Raising or lowering levels of contractility and mitosis rate resulted in non-progressive clefts. We also show that lowered cell-cell adhesion in the cleft region and increased cleft cell-matrix adhesions are required for cleft progression. Using a classifier-based analysis, the relative importance of these four contributing cellular factors for effective cleft progression was determined as follows: cleft cell contractility, cleft region cell-cell adhesion strength, epithelial cell mitosis rate, and cell-matrix adhesion strength.

In this paper, we introduce a biologically motivated dynamic graph-based growth model to describe and predict the stages of cleft formation during the process of branching morphogenesis in the submandibular mouse gland (SMG) from 3 hrs after embryonic day E12 to 8 hrs after embryonic day E12, which can be considered as E12.5. Branching morphogenesis is the process by which many mammalian exocrine and endocrine glands undergo significant morphological transformations, from a primary bud to an adult organ. Although many studies have investigated the cellular and molecular mechanisms driving branching morphogenesis, it is not clear how the shape changes that are inherent to establishing organ structure are produced. Using morphological features extracted from sequential images of SMG organ cultures we were able to develop a dynamic graph-based predictive model that is able to mimic the process of cleft formation and predict the final state. In addition, we compare our model to a state-of-the-art Glazier-Graner-Hogeweg (GGH) simulative tool, and demonstrate that the dynamic graph-based predictive model has comparable accuracy in modeling growth of clefts across SMG developmental stages, as well as faster convergence to the target SMG morphology.

Biomarkers of Mycobacterium tuberculosis complex (MTBC) mutate over time. Among the biomarkers of MTBC, spacer oligonucleotide type (spoligotype) and Mycobacterium Interspersed Repetitive Unit (MIRU) patterns are commonly used to genotype clinical MTBC strains. In this study, we present an evolution model of spoligotype rearrangements using MIRU patterns to disambiguate the ancestors of spoligotypes, in a large patient dataset from the United States Centers for Disease Control and Prevention (CDC). Based on the contiguous deletion assumption and rare observation of convergent evolution, we first generate the most parsimonious forest of spoligotypes, called a spoligoforest, using three genetic distance measures. An analysis of topological attributes of the spoligoforest and number of variations at the direct repeat (DR) locus of each strain reveals interesting properties of deletions in the DR region. First, we compare our mutation model to existing mutation models of spoligotypes and find that our mutation model produces as many within-lineage mutation events as other models, with slightly higher segregation accuracy. Second, based on our mutation model, the number of descendant spoligotypes follows a power law distribution. Third, contrary to prior studies, the power law distribution does not plausibly fit to the mutation length frequency. Finally, the total number of mutation events at consecutive DR loci follows a bimodal distribution, which results in accumulation of shorter deletions in the DR region. The two modes are spacers 13 and 40, which are hotspots for chromosomal rearrangements. The change point in the bimodal distribution is spacer 34, which is absent in most MTBC strains. This bimodal separation results in accumulation of shorter deletions, which explains why a power law distribution is not a plausible fit to the mutation length frequency.

Strains of the Mycobacterium tuberculosis complex (MTBC) can be classified into coherent lineages of similar traits based on their genotype. We present a tensor clustering framework to group MTBC strains into sublineages of the known major lineages based on two biomarkers: spacer oligonucleotide type (spoligotype) and mycobacterial interspersed repetitive units (MIRU). We represent genotype information of MTBC strains in a high-dimensional array in order to include information about spoligotype, MIRU, and their coexistence using multiple-biomarker tensors. We use multiway models to transform this multidimensional data about the MTBC strains into two-dimensional arrays and use the resulting score vectors in a stable partitive clustering algorithm to classify MTBC strains into sublineages. We validate clusterings using cluster stability and accuracy measures, and find stabilities of each cluster. Based on validated clustering results, we present a sublineage structure of MTBC strains and compare it to the sublineage structures of SpolDB4 and MIRU-VNTRplus.

Two-way arrays or matrices are often not enough to represent all the information in the data and standard two-way analysis techniques commonly applied on matrices may fail to find the underlying structures in multi-modal datasets. Multiway data analysis has recently become popular as an exploratory analysis tool in discovering the structures in higher-order datasets, where data have more than two modes. We provide a review of significant contributions in the literature on multiway models, algorithms as well as their applications in diverse disciplines including chemometrics, neuroscience, social network analysis, text mining and computer vision.

Recently the Birkhoff-von Neumann load-balanced (LB) switch has become a promising switch design due to its high scalability properties and simple control. The performance of the LB switch was studied under strong assumptions such as infinite buffers and admissible traffic conditions. However, both such assumptions may be violated in multi-hop networks since admissibility requirement cannot be maintained unless some inter-switch feedback mechanism is implemented, and infinite buffers are not feasible either. This paper considers the performance of the LB switch with finite central stage buffers under both (i) admissible and (ii) inadmissible input traffic conditions. Its contributions are two folds: firstly, by means of mathematical model we demonstrate that the load-balanced switch has a non-zero cell dropping probability due to buffer overflow even under the admissible input traffic assumptions. Secondly, cell loss probabilities are even higher and large buffers are required under inadmissible traffic conditions to cope with such behavior.

With rapid developments in the community mesh networks and wireless sensor networks research, the need for more efficient channel access techniques for multi-hop wireless networks has become eminent. In this work, we propose a novel hybrid channel access scheme that spatially limits the contention in the network such that 2-hop neighbors access the channel contention-free among each other whereas only the immediate neighbors may contend among each other. The contention among neighbors can be handled much more efficiently by a basic CSMA protocol as if operating in a single-hop network. We provide a general framework for the SLICON scheme and compare its performance to the conventional RTS/CTS-based collision avoidance scheme. By a case study using the IEEE 802.11 protocol as the underlying CSMA protocol, the proposed scheme pleads itself as a more efficient alternative to the RTS/CTS based collision avoidance scheme for large and dense multi-hop ad hoc networks with stationary nodes, such as wireless mesh and sensor networks.

The broadcast nature of a wireless link provides a natural eavesdropping and intervention capability to an adversary. Thus, securing a wireless link is essential to the security of a wireless network, and key generation algorithms are necessary for securing wireless links. However, traditional key agreement algorithms can be very costly in many settings, e.g. in wireless ad-hoc networks, since they consume scarce resources such as bandwidth and battery power.

In this paper we define a new metric for quantifying the degree of anonymity collectively afforded to users of an anonymous communication system. We show how our metric, based on the permanent of a matrix, can be useful in evaluating the amount of information needed by an observer to reveal the communication pattern as a whole. We also show how our model can be extended to include probabilistic information learned by an attacker about possible sender-recipient relationships. Our work is intended to serve as a complementary tool to existing information-theoretic metrics, which typically consider the anonymity of the system from the perspective of a single user or message.

The central paradigm of today's successful Internet is to keep the network core simple and move complexity towards the network end points. Unfortunately, this very paradigm limits network management and control capabilities, and creates opportunities for attacks such as worms, viruses, and spam that often seriously disrupt and degrade Internet and user performance. The thrust of this paper is that such problems cannot be effectively solved unless a paradigm shift is adopted. Towards a more secure and manageable Internet, we propose "virtualization" of the Internet, by carefully balancing its scalability and programmability properties. Our objective is to provide a programmable virtual Internet to users and to let them manage, control, and optimize it based on their individual needs.

Secure communications between large number of sensor nodes that are randomly scattered over a hostile territory, necessitate efficient key distribution schemes. However, due to limited resources at sensor nodes such schemes cannot be based on post deployment computations. Instead, pairwise (symmetric) keys are required to be pre-distributed by assigning a list of keys, (a.k.a. key-chain), to each sensor node. If a pair of nodes does not have a common key after deployment then they must find a key-path with secured links. The objective is to minimize the keychain size while (i) maximizing pairwise key sharing probability and resilience, and (ii) minimizing average key-path length. This paper presents a deterministic key distribution scheme based on Expander Graphs. It shows how to map the parameters (e.g., degree, expansion, and diameter) of a Ramanujan Expander Graph to the desired properties of a key distribution scheme for a physical network topology.

This work presents a graph-based representation (a.k.a., cell-graph) of histopathological images for automated cancer diagnosis by probabilistically assigning a link between a pair of cells (or cell clusters). Since the node set of a cell-graph can include a cluster of cells as well as individual ones, it enables working with low-cost, low-magnification photomicrographs. The contributions of this work are twofold. First, it is shown that without establishing a pairwise spatial relation between the cells (i.e., the edges of a cell-graph), neither the spatial distribution of the cells nor the texture analysis of the images yields accurate results for tissue level diagnosis of brain cancer called malignant glioma. Second, this work defines a set of global metrics by processing the entire cell-graph to capture tissue level information coded into the histopathological images. In this work, the results are obtained on the photomicrographs of 646 archival brain biopsy samples of 60 different patients. It is shown that the global metrics of cell-graphs distinguish cancerous tissues from noncancerous ones with high accuracy (at least 99 percent accuracy for healthy tissues with lower cellular density level, and at least 92 percent accuracy for benign tissues with similar high cellular density level such as nonneoplastic reactive/inflammatory conditions).

Distance estimation is important to many Internet applications. It can aid a World Wide Web client when selecting among several potential candidate servers or among candidate peer-to-peer servers. It can also aid in building efficient overlay or peer-to-peer networks that react dynamically to changes in the underlying Internet. One of the approaches to distance (i.e., time delay) estimation in the Internet is based on placing tracer stations in key locations and conducting measurements between them. The tracers construct an approximated map of the Internet after processing the information obtained from these measurements. This work presents a novel algorithm, based on algebraic tools, that computes additional distances, which are not explicitly measured. As such, the algorithm extracts more information from the same amount of measurement data. Our algorithm has several practical impacts. First, it can reduce the number of tracers and measurements without sacrificing information. Second, our algorithm is able to compute distance estimates between locations where tracers cannot be placed. To evaluate the algorithm's performance, we tested it both on randomly generated topologies and on real Internet measurements. Our results show that the algorithm computes up to 50%-200% additional distances beyond the basic tracer-to-tracer measurements.

This paper presents a comprehensive study of the effect of job interleaving by preemption on the throughput of a single server where requests arrive with a given processing time and slack. The problem is to decide which requests to serve so as to maximize the server's utilization. This simple model captures many situations, both at the application (e.g., delivery of video) as well as at the network/transmission levels (e.g., scheduling of packets from input to output interface of a switch). The problem is on-line in nature, and thus we use competitive analysis for measuring the performance of our scheduling algorithms. We consider two modes of operation - with and without commitment - and derive upper and lower bounds for each case. Since competitive analysis is based on the worst-case scenario, the average-case performance of the algorithms is also examined by a simulation study.

A virtual private network (VPN) aims to emulate the services provided by a private network over the shared Internet. The endpoints of a VPN are connected using abstractions such as virtual channels (VCs) of ATM or label switched paths (LSPs) of MPLS technologies. Reliability of an end-to-end VPN connection depends on the reliability of the links and nodes in the fixed path that it traverses in the network. In order to ensure service quality and availability in a VPN, seamless recovery from failures is essential. This work considers the problem of fast recovery in the recently proposed VPN hose model. In the hose model, bandwidth is reserved for traffic aggregates instead of pairwise specifications to allow any traffic pattern among the VPN endpoints. This work assumes that the VPN endpoints are connected using a tree structure and, at any time, at most one tree link can fail (i.e., single link failure model). A restoration algorithm must select a set of backup edges and allocate necessary bandwidth on them in advance, so that the traffic disrupted by failure of a primary edge can be re-routed via backup paths. We aim at designing an optimal restoration algorithm to minimize the total bandwidth reserved on the backup edges. This problem is a variant of optimal graph augmentation problem which is NP-complete. Thus, we present a polynomial-time approximation algorithm that guarantees a solution which is at most 16 times of the optimum. The algorithm is based on designing two reductions to convert the original problem to one of adding minimum cost edges to the VPN tree so that the resulting graph is 2-connected, which can be solved in polynomial time using known algorithms. The two reductions introduce approximation factors of 8 and 2, respectively, thus resulting in a 16-approximation algorithm with polynomial time complexity.

Virtual private networks (VPNs) provide customers with predictable and secure network connections over a shared network. The recently proposed hose model for VPNs allows for greater flexibility since it permits traffic to and from a hose endpoint to be arbitrarily distributed to other endpoints. We develop novel algorithms for provisioning VPNs in the hose model. We connect VPN endpoints using a tree structure and our algorithms attempt to optimize the total bandwidth reserved on edges of the VPN tree. We show that even for the simple scenario in which network links are assumed to have infinite capacity, the general problem of computing the optimal VPN tree is NP-hard. Fortunately, for the special case when the ingress and egress bandwidths for each VPN endpoint are equal, we can devise an algorithm for computing the optimal tree whose time complexity is O(mn), where m and n are the number of links and nodes in the network, respectively. We present a novel integer programming formulation for the general VPN tree computation problem (that is, when ingress and egress bandwidths of VPN endpoints are arbitrary) and develop an algorithm that is based on the primal-dual method. Our experimental results with synthetic network graphs indicate that the VPN trees constructed by our proposed algorithms dramatically reduce bandwidth requirements (in many instances, by more than a factor of 2) compared to scenarios in which Steiner trees are employed to connect VPN endpoints.

Distance estimation is important to many Internet applications, most notably for a WWW client that needs to select a server among several potential candidates. Current approaches to distance (i.e., time delay) estimation in the Internet are based on placing Tracer stations in key locations and conducting measurements between them. The Tracers construct an approximated map of the Internet after processing the information obtained from these measurements. This work presents a novel algorithm, based on algebraic tools, that computes additional distances, which are not explicitly measured. As such, the algorithm extracts more information from the same amount of measurement data. Our algorithm has several practical imparts. First, it can reduce the number of Tracers and measurements without sacrificing information. Second, our algorithm is able to compute distance estimates between locations where Tracers cannot be placed. This is especially important when unidirectional measurements are conducted, since such measurements require specialized equipment which cannot be placed everywhere. To evaluate the algorithm's performance, we tested it both on randomly generated topologies and on real Internet measurements. Our results show that the algorithm computes up to 50-200% additional distances beyond the basic Tracer-to-Tracer measurements.

Virtual Private Networks (VPNs) provide customers with predictable and secure network connections over a shared network. The recently proposed hose model for VPNs allows for greater flexibility since it permits traffic to and from a hose endpoint to be arbitrarily distributed to other endpoints. In this paper, we develop novel algorithms for provisioning VPNs in the hose model. We connect VPN endpoints using a tree structure and our algorithms attempt to optimize the total bandwidth reserved on edges of the VPN tree. We show that even for the simple scenario in which network links are assumed to have infinite capacity, the general problem of computing the optimal VPN tree is NP hard. Fortunately, for the special case when the ingress and egress bandwidths for each VPN endpoint are equal, we can devise an algorithm for computing the optimal tree whose time complexity is O (mn), where m and n are the number of links and nodes in the network, respectively. We present a novel integer programming formulation for the general VPN tree computation problem (that is, when ingress and egress bandwidths of VPN endpoints are arbitrary) and develop an algorithm that is based on the primal-dual method. Our experimental results with synthetic network graphs indicate that the VPN trees constructed by our proposed algorithms dramatically reduce bandwidth requirements (in many instances, by more than a factor of 2) compared to scenarios in which Steiner trees are employed to connect VPN endpoints.

Consider a setting in which a group of nodes, situated in a large underlying network, wishes to reserve bandwidth on which to support communication. Virtual private networks (VPNs) are services that support such a construct; rather than building a new physical network on the group of nodes that must be connected, bandwidth in the underlying network is reserved for communication within the group, forming a virtual “sub-network.”

The mirror sites approach has been proposed previously for reducing the access delay and providing load balancing in network servers. In the mirror site approach a file, such as a multimedia book, is replicated and dispersed over multiple servers and can be requested in parallel. However, to limit the bandwidth waste, each server maintains not the entire file but only a portion of it. Current solutions to provide parallel access to multiple servers are based on breaking the file into b "pieces" using forward error correction (FEC) codes or their variants. In such techniques any k/spl les/b pieces are necessary and sufficient to construct the file. Thus, protection of the file from unauthorized access has to be based on encryption. As a result, as the degree of parallelism increases a trade off occurs between the security overhead and access delay. In this work, we propose new file dispersal and access control protocols to reduce the security overhead significantly. Our protocols are based on the combinatorial techniques which break a file into small pieces in a similar way to FEC code. Thus, at least k pieces are necessary to construct the file. However, in contrast with the previous approaches, not every k pieces are sufficient. Capitalizing on this property this work presents secure dispersal and access protocols that aim to minimize the overhead at the servers.

This paper shows how to provide an adaptive real-time group multicast (many-to-many) communication service. Adaptive means that the number of nodes that transmit to the multicast group is continuously changing. In order to meet deterministic quality-of-service (QoS) requirements of a real-time group multicast, some communication resources are reserved. We show (1) how bandwidth is reserved for each multicast group and (2) how an active source can dynamically share the bandwidth allocated to this multicast group with other active group members. Quality-of-service support for a real-time multicast group is based on time-driven priority. In this scheme the time is divided into time frames of fixed duration, and all the time frames are aligned by using a common global time reference, which can be obtained from the Global Positioning System. Bandwidth is allocated to a multicast group as a whole, rather than individually to each user. The allocation is done by reserving time intervals within time frames in a periodic fashion. This type of allocation raises two problems that are studied in this paper: (1) scheduling: how time intervals are reserved to each multicast group and (2) adaptive sharing: how the active (transmitting) participants can dynamically share the time intervals that have been reserved for their multicast group. The proposed approach is based on the embedding of multiple virtual rings, one for each multicast group. By using the virtual rings, it is simple to route messages to all the participants while minimizing the bound on the buffer sizes and queueing delays. The final part of this paper introduces a scalable growth of the multicast group by adding multiple subtrees to the virtual ring.

This paper presents algorithms, heuristics and lower bounds for an optimal sharing of network resources among several multicast groups that coexist in the network. Group (i.e., many-to-many) multicasting is a demanding service since any member can become a sender independently from the others. We consider a shared tree as the backbone of a group multicasting session. Considering each multicast session in isolation and independently may cause congestion on some links and reduce network utilization. Thus, we define the multicast packing problem in which the network tries to accommodate simultaneously all the multicast groups while trying to avoid bottlenecks on the links for higher throughput (i.e., minimize the maximum link sharing among multicast groups). Minimization of maximum congestion is achieved at the expense of increasing the size of some multicast trees which in turn impacts the delay. This trade-off is addressed by adding a penalty term to the objective function of the optimal packing formulation. The penalty term is a function of the amount of dilation from the size of the optimal tree obtained for each group multicast independently from the others (i.e., in isolation). Since the mathematical programming formulation for the optimization problem is computationally intractable, we resort to suboptimal solutions with heuristics. Our heuristic method aims to reduce the sharing of a link while ensuring that the size of multicast trees will never exceed /spl alpha/OPT/sup k/ where OPT/sup k/ is the size of the optimum tree for multicast group k in isolation. Optimum multicast tree for each group (in isolation) is computed by using cutting-plane inequalities and the branch-and-cut algorithm. In order to evaluate the performance of our approximation, we derive lower bounds on the problem. Our first lower bound on the maximum congestion is a theoretical one and puts a cap on the following two constructive lower bounds. The lower bounds and the heuristic method are implemented and it is shown that the maximum congestion obtained by the heuristic method is quite close to the constructive lower bounds.

This paper presents an extensive performance study in order to identify some tradeoffs between tree-based and ring-based reliable group multicast protocols. The motivation for such a study is the following observation. In communications network routing from one node to another over a tree embedded in the network is intuitively a good strategy, since it typically results in a route length of O(log n) links, while routing from one node to another over a ring embedded in the network would result in route length of O(n) links. However, in a group multicast (many-to-many) the overall number of links traversed by each packet for both tree and ring embedding is typically O(N), so both approaches have similar communication requirements. In reliable group multicast protocols the traffic pattern is complex, since packets are sent from a multicast source to the multiple destinations, and then some control packets are sent back to the source, and this can result in resending of some of the original packets. Consequently, determining under what condition the tree-based approach is better than ring-based approach is not obvious. The key criteria for evaluating the performance of a reliable group multicast protocol is (i) how many successful multicast were achieved per unit time, and (ii) what is the efficiency of the multicast, namely, the ratio between the number of successful transmission and the total number of packets that were transmitted. Under the above criteria it is shown that the ring-based multicast often performs better than the tree-based multicast. One of the main reasons for this result is that ring-based multicast is window-based with simple and effective management of acknowledgments and retransmissions, while the tree-base is rate-based with complex and slow management of acknowledgments and retransmissions.

We evaluate the performance of the copy network under three scenarios: (1) non-buffer non-splitting copy network (NBNS), (2) shared-input buffer-non-splitting copy network (SIBNS), and (3) shared-input-buffer-splitting copy network (SIBS). For NBNS, we derived the exact overflow and cell loss probabilities instead of the Chernoff bound. Furthermore, we propose a general Markov model, a novel theoretical approach, for the performance analysis of the shared-input-buffer copy networks. This analysis method can be applied for both SIBNS and SIBS. In our work, theoretical and simulation results are presented for every scenario. The comparison shows that improved performance of lower cell loss and higher throughput can be achieved by shared-input-buffer (SIBNS and SIBS). However the tradeoff is the long cell delay. With the splitting algorithm, SIBS can provide better performance than the NBNS and SIBNS.

This paper presents algorithms, heuristics and lower bounds addressing optimization issues in many-to-many multicasting. Two main problems are addressed: (1) a precise combinatorial comparison of optimal multicast trees with optimal multicast rings, (2) an optimized sharing of network resources (i.e., nodes and links) among multiple multicast groups that coexist. The former is central to the choice of multicast protocols and their performance, while the latter is crucial for network utilization. The first problem is treated as a comparison of Steiner tree and traveling salesman problems on the same input set. The underlying integer programming problems are solved to optimum by using cutting-plane inequalities and the branch-and-cut algorithm. In addition to these exact solutions, fast heuristics are presented for approximate solutions. The second problem is formulated as a packing problem in which the network tries to accommodate all the multicast groups by optimizing the utilization of resources. Precise mathematical programming formulations, lower bounds and a heuristic for the underlying optimization problem are presented. The heuristic aims to accommodate multiple multicast groups while avoiding bottlenecks on the links for higher throughput. The heuristics and exact algorithms are implemented on various networks and multicast groups. The simulations show that multicast trees can be built by using 25% fewer links than the rings, both for optimal and suboptimal constructions. The packing heuristic is also implemented and its performance is compared to the constructive lower bound.

The routing techniques used in conventional packet radio networks are not suitable for dynamic multihop wireless networks because of their unique architecture. In this paper a new hierarchical multihop routing algorithm is introduced which balances the cost of location-update and path-finding operations by partitioning the terminals and mobile base stations to produce a virtual topology. Based on the virtual topology each network entity stores a fraction of the network topology information and maintains the routing efficiency. Finally, the performance of the hierarchical multihop routing algorithm is investigated through simulations.

This paper shows how to provide an adaptive real-time group multicast (many-to-many) communication service. Such a service can be used by applications, like audio/video tele-conferencing, that require low loss, and bounded delay and jitter. In order to meet deterministic quality of service (QoS) requirements of a real-time group multicast, some communication resources are reserved. In this work we show (i) how bandwidth is reserved for each group, and (ii) how an active user in a multicast group can dynamically share, in an efficient and fair manner, the bandwidth allocated to its group. Quality of service support for a real-time multicast group is based on time driven priority. In this scheme the time is divided into time frames of fixed duration and all the time frames are aligned by using a global time reference which can be obtained from GPS (global positioning system). Bandwidth is allocated to a multicast group as a whole, rather than individually to each user. The allocation is done by reserving time intervals within time frames in some periodic fashion. This sort of allocation raises two problems that are studied in this paper: (1) scheduling: how time intervals are reserved to each multicast group, and (2) adaptive sharing: how the participants dynamically share the time intervals that have been reserved for their multicast group. The proposed approach is based on embedding multiple virtual rings, one for each multicast group. By using the virtual rings it is simple to route messages to all the participants, while minimizing the bound on the buffer sizes and queuing delays.

This paper presents a protocol and design for concurrent and reliable group multicast (many-to-many) from bursty data sources in general networks. In a group multicast, any node can be a multicast source and multiple nodes may start to multicast simultaneously, i.e., an asynchronous access to the network. The reliable multicast protocol presented is window based with a combined sender and receiver initiation of the recovery protocol. In reliable multicasting the necessary requirement is to ensure that data is received correctly by all the active members of the multicast group. The approach taken is to combine the multicast operation with the internal flow control. As a result, it is possible to provide: (1) congestion-free multicast routing with a single and immediate acknowledgment message to the sender. Furthermore, in every multicast, (2) a node can access all the capacity allocated to its group with no delay, however, if several nodes are active in the same group, then the capacity will be shared fairly. In addition, (3) each sender in the multicast group uses a single timer, and (4) a node can become active or inactive in a transparent fashion, i.e., there is no need to explicitly notify the members of the group. A multiple criteria optimization study of the bandwidth allocation to each multicast group is presented. The optimization problem has two min-max objective functions: (1) for delay, which is caused by the number of links needed to connect the group, and (2) for congestion, which is caused by sharing a link among multiple multicast groups. The bandwidth allocation among multicast groups sharing the same link are further optimized using the max-min fairness criterion.

The problem solved in this work is how multiple nodes in a network with an arbitrary topology can broadcast concurrently, in an asynchronous manner, to all other nodes. Asynchronous means that the nodes do not coordinate their broadcast, and, therefore, it is possible that all nodes will start to broadcast at the same time. Simultaneous broadcast by many nodes can cause traffic congestion, which can result in a traffic loss. The main property of the broadcast algorithms presented in this work is that under any arbitrary broadcast pattern there will be no packet or cell loss due to internal traffic congestion. The routing mechanism used by the broadcast algorithm can be viewed as a variant of deflection routing, which means that a node makes on-line routing decisions based on the local flow of traffic (i.e., internal load conditions). Unlike other deflection techniques, the MetaNet routing is along a global sense of direction, which guarantees that packets will reach their destinations. Thus, we call this method convergence routing (previous deflection algorithms did not guarantee deterministic routing convergence, i.e., a cell/packet can be deflected indefinitely inside the network). As a result of the convergence property, the deflection routing used in this work is the only one with broadcast capability.

This paper shows how to optimize the routing decisions in a nondeterministic routing algorithm called convergence routing in which routes may change depending on the traffic conditions. The routing algorithm guarantees a loss-free delivery of data packets from bursty sources, and a deterministic bound on the route length in arbitrary topology networks. The routing decisions are based on assigning routing priorities to the links such that a packet is forwarded to the highest priority link which is not blocked. Routing priorities are assigned using a local-greedy metric which minimizes the distance (number of hops) to the destination. This work shows that routing decisions using a local-greedy metric are not optimal, and the performance of the algorithm can be improved substantially by using new measures. Thus, various look-ahead metrics which take into account the potential gain on the other switching nodes toward the destination of a packet are suggested. The contributions of this work are: (1) a new analytical model to capture the behavior of a switching node; (2) an iterative optimization technique to set routing priorities according to various look-ahead measures; and (3) heuristics to ensure the stability of the routing priorities. The optimization objective is to maximize the throughput by minimizing the maximum total flow carried on a link in the network under static traffic model. The performance is studied computationally on various networks and traffic matrices. It is shown that up to a 50% performance increase can be obtained by optimizing the routing priorities.

This paper presents a new design methodology and tools to construct a packet switched network with bursty data sources. This network design combines two important properties for arbitrary traffic pattern: (1) the aggregate throughput is scalable and (2) there is no packet loss within the subnet. More specifically, given a bounded number of ports in every switching node, the design is based on the construction of multiple virtual rings under the following constraints: (1) the virtual rings are pairwise edge-disjoint and (2) there is at least one virtual ring between any pair of nodes. The target topology is obtained from the edge union of the multiple virtual rings. The two constraints ensure no loss due to congestion inside a network with arbitrary traffic pattern and that packets will reach (or converge) their destinations. The virtual rings are constructed by using combinatorial block designs together with an algorithm for realizing any size networks. It is shown that the bound on the maximum route length, under the two constraints, is O(/spl radic/N) for an N-node network, This sublinear bound facilitates the throughput scalability property.

This paper presents a protocol and design for concurrent and reliable group multicast (many-to-many) from bursty data sources in general networks. In a group multicast, any node can be a multicast source and multiple nodes may start to multicast simultaneously, i.e., an asynchronous access to the network. The reliable multicast protocol presented in this work is window based with a combined sender and receiver initiation of the recovery protocol. In reliable multicasting the necessary requirement is to ensure that data is received correctly by all the active members of the multicast group. The approach taken in this work is to combine the multicast operation with the internal flow control, as a result, it is possible to provide: (1) loss-free multicast routing with a single and immediate acknowledgement message to the sender. Furthermore, in every multicast, (2) a node can access all the capacity allocated to its group with no delay, however, if several nodes are active in the same group then the capacity will be shared fairly. In addition, (3) each sender in the multicast group uses a single timer, and (4) a node can join and leave a multicast group in a transparent fashion, i.e., there is no need to explicitly notify the members of the group. A multiple criteria optimization study of the bandwidth allocation to each multicast group is presented. The optimization problem has two min-max objective functions: (1) for delay, which caused by the number of links needed to connect the group, and (2) for congestion, which is caused by sharing a link among multiple multicast groups. The bandwidth allocation among multicast group sharing the same link are further optimized using the max-min fairness criterion.

The paper presents a new design methodology and tools to construct a switch-based LAN with (i) scalable throughput, (ii) no loss due to congestion, and (iii) two routing modes: FIFO or non-FIFO. More specifically, given a bounded degree (number of switch ports) at each node, the design is based on the construction of multiple virtual rings under the following constraints: (i) the virtual rings are pairwise edge-disjoint, and (ii) there is at least one virtual ring between any pair of nodes. The target topology is obtained from the edge union of the multiple virtual rings. The objectives of the above two constraints are (i) to ensure no loss due to congestion inside the network of bursty traffic sources, and (ii) to ensure convergence of packets/cells to their destinations. The virtual rings are constructed by a new methodology that employs combinatorial block designs together with a new algorithm for realizing any size networks. It is shown that the bound on the maximum route length, under the two constraints, is O(/spl radic/N) for an N-node network.

This paper presents fault-tolerant protocols for fast packet switch networks with convergence routing. The objective is to provide, after a link or a node (switch) failure, fast reconfiguration and continuous host-to-host communication. Convergence routing is a variant of deflection routing, which combines in a dynamic fashion, the on-line routing decision with the traffic load inside the network. Unlike other deflection techniques, convergence routing guarantees that packets will reach or converge to their destinations. This fault-tolerant solution is designed for a switch-based (i.e., arbitrary topology) LAN architecture called MetaNet. The original MetaNet's convergence routing scheme has been modified in order to facilitate the property that the packet header need not be recomputed after a failure and/or a reconfiguration. This is achieved by having, at the network interface, a translator that maps the unique destination address to a virtual address. The virtual addresses are stored at the packet header, and used for convergence routing, with a global sense of direction over (i) a single spanning tree, and (ii) over two edge-disjoint spanning trees, for redundancy (fault tolerance) and greater efficiency.>

This paper considers the combined problem of finding allocation of wavelengths to the stations (configuration) and finding the associated routing of the traffic to minimize congestion (the amount of maximum flow on any link). This work presents efficient algorithms for computing both upper and lower bounds on the congestion. The upper bounds-to obtain approximate solutions of this problem-are based modification of two heuristics (i) variable depth local search, and (ii) simulated annealing. A more significant contribution is a lower bound computation based on building flow trees to find a lower bound on the total flow, and then distributing the total flow over the links provide a lower bound on the congestion. This technique yields a tool which can be used in evaluating the quality of heuristic algorithms, and determining a termination criteria during minimization. This technique can be applied to other problems with flow-based objectives. Performance of the heuristics is analysed, and compared via simulation studies. It is shown that the heuristics perform on the average, within 20% of the computed lower bound, and 15% better than the previous methods to solve this problem.>

Various virtual embedding configurations for the MetaNet architecture are examined. The suggested embedding structures are compared and contrasted, and design parameters and their trade-off are identified. Fairness is used as a mechanism to ensure that the network internally will not be overloaded. Therefore, packets can be sent from source to destination on a route which is close to the shortest path. The routing on the MetaNet is a variant of deflection routing. It makes online routing decisions based on the local flow of traffic (load conditions). MetaNet routing is along a global sense of direction, which guarantees that packets will reach their destinations. This method is called convergence routing. The MetaNet is a LAN/MAN architecture with an "arbitrary topology" (i.e., a switch-based LAN).