Atish Das Sarma
Research Subjects
Atish Das Sarma has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Gopal Pandurangan
University of Houston, Houston, USA
7 shared publications - Anisur Rahaman Molla
Indian Statistical Institute, Kolkata, Kolkata, India
5 shared publications - Amitabh Trehan
Durham University
1 shared publication - Danupon Nanongkai
MPI-INF Germany
1 shared publication - Dennis DeCoste
Machine Learning Group eBay Research Labs 95125 San Jose CA USA
1 shared publication - Eli Upfal
Department of Computer Science Brown University 02912 Providence RI USA
1 shared publication
Latest Publications
- Open Access
Fast Distributed PageRank Computation
Over the last decade, PageRank has gained importance in a wide range of applications and domains, ever since it first proved to be effective in determining node importance in large graphs (and was a pioneering idea behind Google's search engine). In distributed computing alone, PageRank vector, or more generally random walk based quantities have been used for several different applications ranging from determining important nodes, load balancing, search, and identifying connectivity structures. Surprisingly, however, there has been little work towards designing provably efficient fully-distributed algorithms for computing PageRank. The difficulty is that traditional matrix-vector multiplication style iterative methods may not always adapt well to the distributed setting owing to communication bandwidth restrictions and convergence rates. In this paper, we present fast random walk-based distributed algorithms for computing PageRanks in general graphs and prove strong bounds on the round complexity. We first present a distributed algorithm that takes $O\big(\log n/\eps \big)$ rounds with high probability on any graph (directed or undirected), where $n$ is the network size and $\eps$ is the reset probability used in the PageRank computation (typically $\eps$ is a fixed constant). We then present a faster algorithm that takes $O\big(\sqrt{\log n}/\eps \big)$ rounds in undirected graphs. Both of the above algorithms are scalable, as each node sends only small ($\polylog n$) number of bits over each edge per round. To the best of our knowledge, these are the first fully distributed algorithms for computing PageRank vector with provably efficient running time.Open AccessFast Distributed PageRank Computation
Over the last decade, PageRank has gained importance in a wide range of applications and domains, ever since it first proved to be effective in determining node importance in large graphs (and was a pioneering idea behind Google's search engine). In distributed computing alone, PageRank vector, or more generally random walk based quantities have been used for several different applications ranging from determining important nodes, load balancing, search, and identifying connectivity structures. Surprisingly, however, there has been little work towards designing provably efficient fully-distributed algorithms for computing PageRank. The difficulty is that traditional matrix-vector multiplication style iterative methods may not always adapt well to the distributed setting owing to communication bandwidth restrictions and convergence rates. In this paper, we present fast random walk-based distributed algorithms for computing PageRanks in general graphs and prove strong bounds on the round complexity. We first present a distributed algorithm that takes $O\big(\log n/\eps \big)$ rounds with high probability on any graph (directed or undirected), where $n$ is the network size and $\eps$ is the reset probability used in the PageRank computation (typically $\eps$ is a fixed constant). We then present a faster algorithm that takes $O\big(\sqrt{\log n}/\eps \big)$ rounds in undirected graphs. Both of the above algorithms are scalable, as each node sends only small ($\polylog n$) number of bits over each edge per round. To the best of our knowledge, these are the first fully distributed algorithms for computing PageRank vector with provably efficient running time. - 8 May 2015
- Published by Springer Nature in Distributed Computing
- Vol. 28 (5) , 309-320
Efficient distributed computation of distance sketches in networks
- 8 May 2015
- Published by Springer Nature in Distributed Computing
- Vol. 28 (5) , 309-320
Efficient distributed computation of distance sketches in networks
- 1 May 2015
- Published by Elsevier in Theoretical Computer Science
- Vol. 581, 45-66
Distributed computation in dynamic networks via random walks
- 1 May 2015
- Published by Elsevier in Theoretical Computer Science
- Vol. 581, 45-66
Distributed computation in dynamic networks via random walks
- Open Access
Fast Approximate Matching of Cell-Phone Videos for Robust Background Subtraction
We identify a novel instance of the background subtraction problem that focuses on extracting near-field foreground objects captured using handheld cameras. Given two user-generated videos of a scene, one with and the other without the foreground object(s), our goal is to efficiently generate an output video with only the foreground object(s) present in it. We cast this challenge as a spatio-temporal frame matching problem, and propose an efficient solution for it that exploits the temporal smoothness of the video sequences. We present theoretical analyses for the error bounds of our approach, and validate our findings using a detailed set of simulation experiments. Finally, we present the results of our approach tested on multiple real videos captured using handheld cameras, and compare them to several alternate foreground extraction approaches.
Open AccessFast Approximate Matching of Cell-Phone Videos for Robust Background Subtraction
We identify a novel instance of the background subtraction problem that focuses on extracting near-field foreground objects captured using handheld cameras. Given two user-generated videos of a scene, one with and the other without the foreground object(s), our goal is to efficiently generate an output video with only the foreground object(s) present in it. We cast this challenge as a spatio-temporal frame matching problem, and propose an efficient solution for it that exploits the temporal smoothness of the video sequences. We present theoretical analyses for the error bounds of our approach, and validate our findings using a detailed set of simulation experiments. Finally, we present the results of our approach tested on multiple real videos captured using handheld cameras, and compare them to several alternate foreground extraction approaches.
- Open Access
Distributed Computation of Sparse Cuts
Finding sparse cuts is an important tool in analyzing large-scale distributed networks such as the Internet and Peer-to-Peer networks, as well as large-scale graphs such as the web graph, online social communities, and VLSI circuits. In distributed communication networks, they are useful for topology maintenance and for designing better search and routing algorithms. In this paper, we focus on developing fast distributed algorithms for computing sparse cuts in networks. Given an undirected $n$-node network $G$ with conductance $\phi$, the goal is to find a cut set whose conductance is close to $\phi$. We present two distributed algorithms that find a cut set with sparsity $\tilde O(\sqrt{\phi})$ ($\tilde{O}$ hides $\polylog{n}$ factors). Both our algorithms work in the CONGEST distributed computing model and output a cut of conductance at most $\tilde O(\sqrt{\phi})$ with high probability, in $\tilde O(\frac{1}{b}(\frac{1}{\phi} + n))$ rounds, where $b$ is balance of the cut of given conductance. In particular, to find a sparse cut of constant balance, our algorithms take $\tilde O(\frac{1}{\phi} + n)$ rounds. Our algorithms can also be used to output a {\em local} cluster, i.e., a subset of vertices near a given source node, and whose conductance is within a quadratic factor of the best possible cluster around the specified node. Both our distributed algorithm can work without knowledge of the optimal $\phi$ value and hence can be used to find approximate conductance values both globally and with respect to a given source node. We also give a lower bound on the time needed for any distributed algorithm to compute any non-trivial sparse cut --- any distributed approximation algorithm (for any non-trivial approximation ratio) for computing sparsest cut will take $\tilde \Omega(\sqrt{n} + D)$ rounds, where $D$ is the diameter of the graph.Open AccessDistributed Computation of Sparse Cuts
Finding sparse cuts is an important tool in analyzing large-scale distributed networks such as the Internet and Peer-to-Peer networks, as well as large-scale graphs such as the web graph, online social communities, and VLSI circuits. In distributed communication networks, they are useful for topology maintenance and for designing better search and routing algorithms. In this paper, we focus on developing fast distributed algorithms for computing sparse cuts in networks. Given an undirected $n$-node network $G$ with conductance $\phi$, the goal is to find a cut set whose conductance is close to $\phi$. We present two distributed algorithms that find a cut set with sparsity $\tilde O(\sqrt{\phi})$ ($\tilde{O}$ hides $\polylog{n}$ factors). Both our algorithms work in the CONGEST distributed computing model and output a cut of conductance at most $\tilde O(\sqrt{\phi})$ with high probability, in $\tilde O(\frac{1}{b}(\frac{1}{\phi} + n))$ rounds, where $b$ is balance of the cut of given conductance. In particular, to find a sparse cut of constant balance, our algorithms take $\tilde O(\frac{1}{\phi} + n)$ rounds. Our algorithms can also be used to output a {\em local} cluster, i.e., a subset of vertices near a given source node, and whose conductance is within a quadratic factor of the best possible cluster around the specified node. Both our distributed algorithm can work without knowledge of the optimal $\phi$ value and hence can be used to find approximate conductance values both globally and with respect to a given source node. We also give a lower bound on the time needed for any distributed algorithm to compute any non-trivial sparse cut --- any distributed approximation algorithm (for any non-trivial approximation ratio) for computing sparsest cut will take $\tilde \Omega(\sqrt{n} + D)$ rounds, where $D$ is the diameter of the graph. - Open Access
Distributed Random Walks
Performing random walks in networks is a fundamental primitive that has found applications in many areas of computer science, including distributed computing. In this paper, we focus on the problem of sampling random walks efficiently in a distributed network and its applications. Given bandwidth constraints, the goal is to minimize the number of rounds required to obtain random walk samples. All previous algorithms that compute a random walk sample of length $\ell$ as a subroutine always do so naively, i.e., in $O(\ell)$ rounds. The main contribution of this paper is a fast distributed algorithm for performing random walks. We present a sublinear time distributed algorithm for performing random walks whose time complexity is sublinear in the length of the walk. Our algorithm performs a random walk of length $\ell$ in $\tilde{O}(\sqrt{\ell D})$ rounds ($\tilde{O}$ hides $\polylog{n}$ factors where $n$ is the number of nodes in the network) with high probability on an undirected network, where $D$ is the diameter of the network. For small diameter graphs, this is a significant improvement over the naive $O(\ell)$ bound. Furthermore, our algorithm is optimal within a poly-logarithmic factor as there exists a matching lower bound [Nanongkai et al. PODC 2011]. We further extend our algorithms to efficiently perform $k$ independent random walks in $\tilde{O}(\sqrt{k\ell D} + k)$ rounds. We also show that our algorithm can be applied to speedup the more general Metropolis-Hastings sampling. Our random walk algorithms can be used to speed up distributed algorithms in applications that use random walks as a subroutine, such as computing a random spanning tree and estimating mixing time and related parameters. Our algorithm is fully decentralized and can serve as a building block in the design of topologically-aware networks.Open AccessDistributed Random Walks
Performing random walks in networks is a fundamental primitive that has found applications in many areas of computer science, including distributed computing. In this paper, we focus on the problem of sampling random walks efficiently in a distributed network and its applications. Given bandwidth constraints, the goal is to minimize the number of rounds required to obtain random walk samples. All previous algorithms that compute a random walk sample of length $\ell$ as a subroutine always do so naively, i.e., in $O(\ell)$ rounds. The main contribution of this paper is a fast distributed algorithm for performing random walks. We present a sublinear time distributed algorithm for performing random walks whose time complexity is sublinear in the length of the walk. Our algorithm performs a random walk of length $\ell$ in $\tilde{O}(\sqrt{\ell D})$ rounds ($\tilde{O}$ hides $\polylog{n}$ factors where $n$ is the number of nodes in the network) with high probability on an undirected network, where $D$ is the diameter of the network. For small diameter graphs, this is a significant improvement over the naive $O(\ell)$ bound. Furthermore, our algorithm is optimal within a poly-logarithmic factor as there exists a matching lower bound [Nanongkai et al. PODC 2011]. We further extend our algorithms to efficiently perform $k$ independent random walks in $\tilde{O}(\sqrt{k\ell D} + k)$ rounds. We also show that our algorithm can be applied to speedup the more general Metropolis-Hastings sampling. Our random walk algorithms can be used to speed up distributed algorithms in applications that use random walks as a subroutine, such as computing a random spanning tree and estimating mixing time and related parameters. Our algorithm is fully decentralized and can serve as a building block in the design of topologically-aware networks. - Open Access
Fast Distributed Computation in Dynamic Networks via Random Walks
The paper investigates efficient distributed computation in dynamic networks in which the network topology changes (arbitrarily) from round to round. Our first contribution is a rigorous framework for design and analysis of distributed random walk algorithms in dynamic networks. We then develop a fast distributed random walk based algorithm that runs in $\tilde{O}(\sqrt{\tau \Phi})$ rounds (with high probability), where $\tau$ is the dynamic mixing time and $\Phi$ is the dynamic diameter of the network respectively, and returns a sample close to a suitably defined stationary distribution of the dynamic network. We also apply our fast random walk algorithm to devise fast distributed algorithms for two key problems, namely, information dissemination and decentralized computation of spectral properties in a dynamic network. Our next contribution is a fast distributed algorithm for the fundamental problem of information dissemination (also called as gossip) in a dynamic network. In gossip, or more generally, $k$-gossip, there are $k$ pieces of information (or tokens) that are initially present in some nodes and the problem is to disseminate the $k$ tokens to all nodes. We present a random-walk based algorithm that runs in $\tilde{O}(\min\{n^{1/3}k^{2/3}(\tau \Phi)^{1/3}, nk\})$ rounds (with high probability). To the best of our knowledge, this is the first $o(nk)$-time fully-distributed token forwarding algorithm that improves over the previous-best $O(nk)$ round distributed algorithm [Kuhn et al., STOC 2010], although in an oblivious adversary model. Our final contribution is a simple and fast distributed algorithm for estimating the dynamic mixing time and related spectral properties of the underlying dynamic network.Open AccessFast Distributed Computation in Dynamic Networks via Random Walks
The paper investigates efficient distributed computation in dynamic networks in which the network topology changes (arbitrarily) from round to round. Our first contribution is a rigorous framework for design and analysis of distributed random walk algorithms in dynamic networks. We then develop a fast distributed random walk based algorithm that runs in $\tilde{O}(\sqrt{\tau \Phi})$ rounds (with high probability), where $\tau$ is the dynamic mixing time and $\Phi$ is the dynamic diameter of the network respectively, and returns a sample close to a suitably defined stationary distribution of the dynamic network. We also apply our fast random walk algorithm to devise fast distributed algorithms for two key problems, namely, information dissemination and decentralized computation of spectral properties in a dynamic network. Our next contribution is a fast distributed algorithm for the fundamental problem of information dissemination (also called as gossip) in a dynamic network. In gossip, or more generally, $k$-gossip, there are $k$ pieces of information (or tokens) that are initially present in some nodes and the problem is to disseminate the $k$ tokens to all nodes. We present a random-walk based algorithm that runs in $\tilde{O}(\min\{n^{1/3}k^{2/3}(\tau \Phi)^{1/3}, nk\})$ rounds (with high probability). To the best of our knowledge, this is the first $o(nk)$-time fully-distributed token forwarding algorithm that improves over the previous-best $O(nk)$ round distributed algorithm [Kuhn et al., STOC 2010], although in an oblivious adversary model. Our final contribution is a simple and fast distributed algorithm for estimating the dynamic mixing time and related spectral properties of the underlying dynamic network. - Open Access
Near-Optimal Random Walk Sampling in Distributed Networks
Performing random walks in networks is a fundamental primitive that has found numerous applications in communication networks such as token management, load balancing, network topology discovery and construction, search, and peer-to-peer membership management. While several such algorithms are ubiquitous, and use numerous random walk samples, the walks themselves have always been performed naively. In this paper, we focus on the problem of performing random walk sampling efficiently in a distributed network. Given bandwidth constraints, the goal is to minimize the number of rounds and messages required to obtain several random walk samples in a continuous online fashion. We present the first round and message optimal distributed algorithms that present a significant improvement on all previous approaches. The theoretical analysis and comprehensive experimental evaluation of our algorithms show that they perform very well in different types of networks of differing topologies. In particular, our results show how several random walks can be performed continuously (when source nodes are provided only at runtime, i.e., online), such that each walk of length $\ell$ can be performed exactly in just $\tilde{O}(\sqrt{\ell D})$ rounds, (where $D$ is the diameter of the network), and $O(\ell)$ messages. This significantly improves upon both, the naive technique that requires $O(\ell)$ rounds and $O(\ell)$ messages, and the sophisticated algorithm of [DasSarma et al. PODC 2010] that has the same round complexity as this paper but requires $\Omega(m\sqrt{\ell})$ messages (where $m$ is the number of edges in the network). Our theoretical results are corroborated through extensive experiments on various topological data sets. Our algorithms are fully decentralized, lightweight, and easily implementable, and can serve as building blocks in the design of topologically-aware networks.Open AccessNear-Optimal Random Walk Sampling in Distributed Networks
Performing random walks in networks is a fundamental primitive that has found numerous applications in communication networks such as token management, load balancing, network topology discovery and construction, search, and peer-to-peer membership management. While several such algorithms are ubiquitous, and use numerous random walk samples, the walks themselves have always been performed naively. In this paper, we focus on the problem of performing random walk sampling efficiently in a distributed network. Given bandwidth constraints, the goal is to minimize the number of rounds and messages required to obtain several random walk samples in a continuous online fashion. We present the first round and message optimal distributed algorithms that present a significant improvement on all previous approaches. The theoretical analysis and comprehensive experimental evaluation of our algorithms show that they perform very well in different types of networks of differing topologies. In particular, our results show how several random walks can be performed continuously (when source nodes are provided only at runtime, i.e., online), such that each walk of length $\ell$ can be performed exactly in just $\tilde{O}(\sqrt{\ell D})$ rounds, (where $D$ is the diameter of the network), and $O(\ell)$ messages. This significantly improves upon both, the naive technique that requires $O(\ell)$ rounds and $O(\ell)$ messages, and the sophisticated algorithm of [DasSarma et al. PODC 2010] that has the same round complexity as this paper but requires $\Omega(m\sqrt{\ell})$ messages (where $m$ is the number of edges in the network). Our theoretical results are corroborated through extensive experiments on various topological data sets. Our algorithms are fully decentralized, lightweight, and easily implementable, and can serve as building blocks in the design of topologically-aware networks. - Open Access
Edge-preserving self-healing: keeping network backbones densely connected
Healing algorithms play a crucial part in distributed P2P networks where failures occur continuously and frequently. Several self-healing algorithms have been suggested recently [IPDPS'08, PODC'08, PODC'09, PODC'11] in a line of work that has yielded gradual improvements in the properties ensured on the graph. This work motivates a strong general phenomenon of edge-preserving healing that aims at obtaining self-healing algorithms with the constraint that all original edges in the graph (not deleted by the adversary), be retained in every intermediate graph. The previous algorithms, in their nascent form, are not explicitly edge preserving. In this paper, we show they can be suitably modified (We introduce Xheal+, an edge-preserving version of Xheal[PODC'11]). Towards this end, we present a general self-healing model that unifies the previous models. The main contribution of this paper is not in the technical complexity, rather in the simplicity with which the edge-preserving property can be ensured and the message that this is a crucial property with several benefits. In particular, we highlight this by showing that, almost as an immediate corollary, subgraph densities are preserved or increased. Maintaining density is a notion motivated by the fact that in certain distributed networks, certain nodes may require and initially have a larger number of inter-connections. It is vital that a healing algorithm, even amidst failures, respect these requirements. Our suggested modifications yield such subgraph density preservation as a by product. In addition, edge preservation helps maintain any subgraph induced property that is monotonic. Also, algorithms that are edge-preserving require minimal alteration of edges which can be an expensive cost in healing - something that has not been modeled in any of the past work.
Open AccessEdge-preserving self-healing: keeping network backbones densely connected
Healing algorithms play a crucial part in distributed P2P networks where failures occur continuously and frequently. Several self-healing algorithms have been suggested recently [IPDPS'08, PODC'08, PODC'09, PODC'11] in a line of work that has yielded gradual improvements in the properties ensured on the graph. This work motivates a strong general phenomenon of edge-preserving healing that aims at obtaining self-healing algorithms with the constraint that all original edges in the graph (not deleted by the adversary), be retained in every intermediate graph. The previous algorithms, in their nascent form, are not explicitly edge preserving. In this paper, we show they can be suitably modified (We introduce Xheal+, an edge-preserving version of Xheal[PODC'11]). Towards this end, we present a general self-healing model that unifies the previous models. The main contribution of this paper is not in the technical complexity, rather in the simplicity with which the edge-preserving property can be ensured and the message that this is a crucial property with several benefits. In particular, we highlight this by showing that, almost as an immediate corollary, subgraph densities are preserved or increased. Maintaining density is a notion motivated by the fact that in certain distributed networks, certain nodes may require and initially have a larger number of inter-connections. It is vital that a healing algorithm, even amidst failures, respect these requirements. Our suggested modifications yield such subgraph density preservation as a by product. In addition, edge preservation helps maintain any subgraph induced property that is monotonic. Also, algorithms that are edge-preserving require minimal alteration of edges which can be an expensive cost in healing - something that has not been modeled in any of the past work.