Searcharxiv⌕ Search

arXiv subjects

Luca Trevisan

Publications and source records attributed to Luca Trevisan.

51 records · Page 3Linked to original sources
arXiv2013

Improved ARV Rounding in Small-set Expanders and Graphs of Bounded Threshold Rank

We prove a structure theorem for the feasible solutions of the Arora-Rao-Vazirani SDP relaxation on low threshold rank graphs and on small-set expanders. We show that if G is a graph of bounded threshold rank or a small-set expander, then an optimal solution of the Arora-Rao-Vazirani relaxation (or of any stronger version of it) can be almost entirely covered by a small number of balls of bounded radius. Then, we show that, if k is the number of balls, a solution of this form can be rounded with an approximation factor of O(sqrt {log k}) in the case of the Arora-Rao-Vazirani relaxation, and with a constant-factor approximation in the case of the k-th round of the Sherali-Adams hierarchy starting at the Arora-Rao-Vazirani relaxation. The structure theorem and the rounding scheme combine to prove the following result, where G=(V,E) is a graph of expansion ϕ(G), λ_k is the k-th smallest eigenvalue of the normalized Laplacian of G, and ϕ_k(G) = \min_{disjoint S_1,...,S_k} \max_{1 <= i <= k} ϕ(S_i) is the largest expansion of any k disjoint subsets of V: if either λ_k >> log^{2.5} k \cdot phi(G) or ϕ_{k} (G) >> log k \cdot sqrt{log n}\cdot loglog n\cdot ϕ(G), then the Arora-Rao-Vazirani relaxation can be rounded in polynomial time with an approximation ratio O(sqrt{log k}). Stronger approximation guarantees are achievable in time exponential in k via relaxations in the Lasserre hierarchy. Guruswami and Sinop [GS13] and Arora, Ge and Sinop [AGS13] prove that 1+eps approximation is achievable in time 2^{O(k)} poly(n) if either λ_k > ϕ(G)/ poly(eps), or if SSE_{n/k} > sqrt{log k log n} \cdot ϕ(G)/ poly(eps), where SSE_s is the minimal expansion of sets of size at most s.

cs.DS↗
arXiv2013

Is Cheeger-type Approximation Possible for Nonuniform Sparsest Cut?

In the {\em nonuniform sparsest cut} problem, given two undirected graphs $G$ and $H$ over the same set of vertices $V$, we want to find a cut $(S,V-S)$ that minimizes the ratio between the fraction of $G$-edges that are cut and the fraction of $H$-edges that are cut. The ratio (which is at most 1 in an optimal solution) is called the {\em sparsity} of the cut. In the {\em uniform sparsest cut} problem, $H$ is a clique over $V$. If $G$ is regular, it is possible to find a solution to the uniform sparsest cut of cost $O(\sqrt{opt})$ in nearly linear time. Is such an approximation, which we call "Cheege-type" approximation, achievable in the non-uniform case? We show that the answer is negative, assuming the Unique Games Conjecture, for general H. Furthermore, the Leighton-Rao linear programming relaxation and the spectral relaxation fail to find such an approximation even if $H$ is a clique over a subset of vertices. Using semidefinite programming, however, we can find Cheeger-type approximations in polynomial time whenever the adjacency matrix of $H$ has rank 1. (This includes the cases in which $H$ is a clique over a subset of vertices.)

cs.DS↗
arXiv2013

Improved Cheeger's Inequality: Analysis of Spectral Partitioning Algorithms through Higher Order Spectral Gap

Let ϕ(G) be the minimum conductance of an undirected graph G, and let 0=λ_1 <= λ_2 <=... <= λ_n <= 2 be the eigenvalues of the normalized Laplacian matrix of G. We prove that for any graph G and any k >= 2, ϕ(G) = O(k) λ_2 / \sqrt{λ_k}, and this performance guarantee is achieved by the spectral partitioning algorithm. This improves Cheeger's inequality, and the bound is optimal up to a constant factor for any k. Our result shows that the spectral partitioning algorithm is a constant factor approximation algorithm for finding a sparse cut if λ_k$ is a constant for some constant k. This provides some theoretical justification to its empirical performance in image segmentation and clustering problems. We extend the analysis to other graph partitioning problems, including multi-way partition, balanced separator, and maximum cut.

cs.DS↗
arXiv2012

A New Regularity Lemma and Faster Approximation Algorithms for Low Threshold Rank Graphs

Kolla and Tulsiani [KT07,Kolla11} and Arora, Barak and Steurer [ABS10] introduced the technique of subspace enumeration, which gives approximation algorithms for graph problems such as unique games and small set expansion; the running time of such algorithms is exponential in the threshold-rank of the graph. Guruswami and Sinop [GS11,GS12], and Barak, Raghavendra, and Steurer [BRS11] developed an alternative approach to the design of approximation algorithms for graphs of bounded threshold-rank, based on semidefinite programming relaxations in the Lassere hierarchy and on novel rounding techniques. These algorithms are faster than the ones based on subspace enumeration and work on a broad class of problems. In this paper we develop a third approach to the design of such algorithms. We show, constructively, that graphs of bounded threshold-rank satisfy a weak Szemeredi regularity lemma analogous to the one proved by Frieze and Kannan [FK99] for dense graphs. The existence of efficient approximation algorithms is then a consequence of the regularity lemma, as shown by Frieze and Kannan. Applying our method to the Max Cut problem, we devise an algorithm that is faster than all previous algorithms, and is easier to describe and analyze.

cs.DS↗
arXiv2012

Approximating the Expansion Profile and Almost Optimal Local Graph Clustering

Spectral partitioning is a simple, nearly-linear time, algorithm to find sparse cuts, and the Cheeger inequalities provide a worst-case guarantee for the quality of the approximation found by the algorithm. Local graph partitioning algorithms [ST08,ACL06,AP09] run in time that is nearly linear in the size of the output set, and their approximation guarantee is worse than the guarantee provided by the Cheeger inequalities by a polylogarithmic $\log^{Ω(1)} n$ factor. It has been a long standing open problem to design a local graph clustering algorithm with an approximation guarantee close to the guarantee of the Cheeger inequalities and with a running time nearly linear in the size of the output. In this paper we solve this problem; we design an algorithm with the same guarantee (up to a constant factor) as the Cheeger inequality, that runs in time slightly super linear in the size of the output. This is the first sublinear (in the size of the input) time algorithm with almost the same guarantee as the Cheeger's inequality. As a byproduct of our results, we prove a bicriteria approximation algorithm for the expansion profile of any graph. Let $ϕ(γ) = \min_{μ(S) \leq γ}ϕ(S)$. There is a polynomial time algorithm that, for any $γ,ε>0$, finds a set $S$ of measure $μ(S)\leq 2γ^{1+ε}$, and expansion $ϕ(S)\leq \sqrt{2ϕ(γ)/ε}$. Our proof techniques also provide a simpler proof of the structural result of Arora, Barak, Steurer [ABS10], that can be applied to irregular graphs. Our main technical tool is that for any set $S$ of vertices of a graph, a lazy $t$-step random walk started from a randomly chosen vertex of $S$, will remain entirely inside $S$ with probability at least $(1-ϕ(S)/2)^t$. This itself provides a new lower bound to the uniform mixing time of any finite states reversible markov chain.

cs.DS↗
arXiv2012

Better Pseudorandom Generators from Milder Pseudorandom Restrictions

We present an iterative approach to constructing pseudorandom generators, based on the repeated application of mild pseudorandom restrictions. We use this template to construct pseudorandom generators for combinatorial rectangles and read-once CNFs and a hitting set generator for width-3 branching programs, all of which achieve near-optimal seed-length even in the low-error regime: We get seed-length O(log (n/epsilon)) for error epsilon. Previously, only constructions with seed-length O(\log^{3/2} n) or O(\log^2 n) were known for these classes with polynomially small error. The (pseudo)random restrictions we use are milder than those typically used for proving circuit lower bounds in that we only set a constant fraction of the bits at a time. While such restrictions do not simplify the functions drastically, we show that they can be derandomized using small-bias spaces.

cs.CC↗
arXiv2012

Information Spreading in Dynamic Graphs

We present a general approach to study the flooding time (a measure of how fast information spreads) in dynamic graphs (graphs whose topology changes with time according to a random process). We consider arbitrary converging Markovian dynamic graph process, that is, processes in which the topology of the graph at time $t$ depends only on its topology at time $t-1$ and which have a unique stationary distribution. The most well studied models of dynamic graphs are all Markovian and converging. Under general conditions, we bound the flooding time in terms of the mixing time of the dynamic graph process. We recover, as special cases of our result, bounds on the flooding time for the \emph{random trip} model and the \emph{random path} models; previous analysis techniques provided bounds only in restricted settings for such models. Our result also provides the first bound for the \emph{random waypoint} model (which is tight for certain ranges of parameters) whose analysis had been an important open question.

cs.DM↗
arXiv2011

A Higher-Order Cheeger's Inequality

A basic fact in algebraic graph theory is that the number of connected components in an undirected graph is equal to the multiplicity of the eigenvalue 1 in the normalized adjacency matrix of the graph. In particular, the graph is disconnected if and only if there are at least two eigenvalues equal to 1. Cheeger's inequality provides an "approximate" version of the latter fact, and it states that a graph has a sparse cut (it is "almost disconnected") if and only if there are at least two eigenvalues that are close to one. It has been conjectured that an analogous characterization holds for higher multiplicities, that is there are $k$ eigenvalues close to 1 if and only if the vertex set can be partitioned into $k$ subsets, each defining a sparse cut. In this paper we resolve this conjecture. Our result provides a theoretical justification for clustering algorithms that use the top $k$ eigenvector to embed the vertices into $\R^k$, and then apply geometric considerations to the embedding.

cs.DS↗
arXiv2008

Max Cut and the Smallest Eigenvalue

We describe a new approximation algorithm for Max Cut. Our algorithm runs in $\tilde O(n^2)$ time, where $n$ is the number of vertices, and achieves an approximation ratio of $.531$. On instances in which an optimal solution cuts a $1-ε$ fraction of edges, our algorithm finds a solution that cuts a $1-4\sqrtε + 8ε-o(1)$ fraction of edges. Our main result is a variant of spectral partitioning, which can be implemented in nearly linear time. Given a graph in which the Max Cut optimum is a $1-ε$ fraction of edges, our spectral partitioning algorithm finds a set $S$ of vertices and a bipartition $L,R=S-L$ of $S$ such that at least a $1-O(\sqrt ε)$ fraction of the edges incident on $S$ have one endpoint in $L$ and one endpoint in $R$. (This can be seen as an analog of Cheeger's inequality for the smallest eigenvalue of the adjacency matrix of a graph.) Iterating this procedure yields the approximation results stated above. A different, more complicated, variant of spectral partitioning leads to an $\tilde O(n^3)$ time algorithm that cuts $1/2 + e^{-Ω(1/\eps)}$ fraction of edges in graphs in which the optimum is $1/2 + ε$.

cs.DS↗
arXiv2008

New Proofs of the Green-Tao-Ziegler Dense Model Theorem: An Exposition

Green, Tao and Ziegler prove ``Dense Model Theorems'' of the following form: if R is a (possibly very sparse) pseudorandom subset of set X, and D is a dense subset of R, then D may be modeled by a set M whose density inside X is approximately the same as the density of D in R. More generally, they show that a function that is majorized by a pseudorandom measure can be written as a sum of a bounded function having the same expectation plus a function that is ``indistinguishable from zero.'' This theorem plays a key role in the proof of the Green-Tao Theorem that the primes contain arbitrarily long arithmetic progressions. In this note, we present a new proof of the Green-Tao-Ziegler Dense Model Theorem, which was discovered independently by ourselves and Gowers. We refer to our full paper for variants of the result with connections and applications to computational complexity theory, and to Gowers' paper for applications of the proof technique to ``decomposition, ``structure,'' and ``transference'' theorems in arithmetic and extremal combinatorics (as well as a broader survey of such theorems).

math.CO↗
arXiv2006

Pseudorandomness and Combinatorial Constructions

In combinatorics, the probabilistic method is a very powerful tool to prove the existence of combinatorial objects with interesting and useful properties. Explicit constructions of objects with such properties are often very difficult, or unknown. In computer science, probabilistic algorithms are sometimes simpler and more efficient than the best known deterministic algorithms for the same problem. Despite this evidence for the power of random choices, the computational theory of pseudorandomness shows that, under certain complexity-theoretic assumptions, every probabilistic algorithm has an efficient deterministic simulation and a large class of applications of the the probabilistic method can be converted into explicit constructions. In this survey paper we describe connections between the conditional ``derandomization'' results of the computational theory of pseudorandomness and unconditional explicit constructions of certain combinatorial objects such as error-correcting codes and ``randomness extractors.''

cs.CC↗
arXiv2005

Gowers Uniformity, Influence of Variables, and PCPs

Gowers introduced, for d\geq 1, the notion of dimension-d uniformity U^d(f) of a function f: G -> \C, where G is a finite abelian group and \C are the complex numbers. Roughly speaking, if U^d(f) is small, then f has certain "pseudorandomness" properties. We prove the following property of functions with large U^d(f). Write G=G_1 x >... x G_n as a product of groups. If a bounded balanced function f:G_1 x ... x G_n -> \C is such that U^{d} (f) > epsilon, then one of the coordinates of f has influence at least epsilon/2^{O(d)}. The Gowers inner product of a collection of functions is a related notion of pseudorandomness. We prove that if a collection of bounded functions has large Gowers inner product, and at least one function in the collection is balanced, then there is a variable that has high influence for at least four of the functions in the collection. Finally, we relate the acceptance probability of the "hypergraph long-code test" proposed by Samorodnitsky and Trevisan to the Gowers inner product of the functions being tested and we deduce applications to the construction of Probabilistically Checkable Proofs and to hardness of approximation.

math.CO↗
arXiv2004

Some Applications of Coding Theory in Computational Complexity

Error-correcting codes and related combinatorial constructs play an important role in several recent (and old) results in computational complexity theory. In this paper we survey results on locally-testable and locally-decodable error-correcting codes, and their applications to complexity theory and to cryptography. Locally decodable codes are error-correcting codes with sub-linear time error-correcting algorithms. They are related to private information retrieval (a type of cryptographic protocol), and they are used in average-case complexity and to construct ``hard-core predicates'' for one-way permutations. Locally testable codes are error-correcting codes with sub-linear time error-detection algorithms, and they are the combinatorial core of probabilistically checkable proofs.

cs.CC↗