arXiv subjects

Ely Porat

Publications and source records attributed to Ely Porat.

At least 55 records · Page 3Linked to original sources

arXiv2012

Feasible Sampling of Non-strict Turnstile Data Streams

We present the first feasible method for sampling a dynamic data stream with deletions, where the sample consists of pairs $(k,C_k)$ of a value $k$ and its exact total count $C_k$. Our algorithms are for both Strict Turnstile data streams and the most general Non-strict Turnstile data streams, where each element may have a negative total count. Our method improves by an order of magnitude the known processing time of each element in the stream, which is extremely crucial for data stream applications. For example, for a sample of size $O(ε^{-2} \log{(1/δ)})$ in Non-strict streams, our solution requires $O((\log\log(1/ε))^2 + (\log\log(1/δ)) ^ 2)$ operations per stream element, whereas the best previous solution requires $O(ε^{-2} \log^2(1/δ))$ evaluations of a fully independent hash function per element. Here $1-δ$ is the success probability and $ε$ is the additive approximation error. We achieve this improvement by constructing a single data structure from which multiple elements can be extracted with very high success probability. The sample we generate is useful for calculating both forward and inverse distribution statistics, within an additive error, with provable guarantees on the success probability. Furthermore, our algorithms can run on distributed systems and extract statistics on the union or difference between data streams. They can be used to calculate the Jaccard similarity coefficient as well.

cs.DS↗

arXiv2012

Pattern Matching in Multiple Streams

We investigate the problem of deterministic pattern matching in multiple streams. In this model, one symbol arrives at a time and is associated with one of s streaming texts. The task at each time step is to report if there is a new match between a fixed pattern of length m and a newly updated stream. As is usual in the streaming context, the goal is to use as little space as possible while still reporting matches quickly. We give almost matching upper and lower space bounds for three distinct pattern matching problems. For exact matching we show that the problem can be solved in constant time per arriving symbol and O(m+s) words of space. For the k-mismatch and k-difference problems we give O(k) time solutions that require O(m+ks) words of space. In all three cases we also give space lower bounds which show our methods are optimal up to a single logarithmic factor. Finally we set out a number of open problems related to this new model for pattern matching.

cs.DS↗

arXiv2012

Worst-case Optimal Join Algorithms

Efficient join processing is one of the most fundamental and well-studied tasks in database research. In this work, we examine algorithms for natural join queries over many relations and describe a novel algorithm to process these queries optimally in terms of worst-case data complexity. Our result builds on recent work by Atserias, Grohe, and Marx, who gave bounds on the size of a full conjunctive query in terms of the sizes of the individual relations in the body of the query. These bounds, however, are not constructive: they rely on Shearer's entropy inequality which is information-theoretic. Thus, the previous results leave open the question of whether there exist algorithms whose running time achieve these optimal bounds. An answer to this question may be interesting to database practice, as it is known that any algorithm based on the traditional select-project-join style plans typically employed in an RDBMS are asymptotically slower than the optimal for some queries. We construct an algorithm whose running time is worst-case optimal for all natural join queries. Our result may be of independent interest, as our algorithm also yields a constructive proof of the general fractional cover bound by Atserias, Grohe, and Marx without using Shearer's inequality. This bound implies two famous inequalities in geometry: the Loomis-Whitney inequality and the Bollobás-Thomason inequality. Hence, our results algorithmically prove these inequalities as well. Finally, we discuss how our algorithm can be used to compute a relaxed notion of joins.

cs.DB↗

arXiv2011

A cuckoo hashing variant with improved memory utilization and insertion time

Cuckoo hashing [4] is a multiple choice hashing scheme in which each item can be placed in multiple locations, and collisions are resolved by moving items to their alternative locations. In the classical implementation of two-way cuckoo hashing, the memory is partitioned into contiguous disjoint fixed-size buckets. Each item is hashed to two buckets, and may be stored in any of the positions within those buckets. Ref. [2] analyzed a variation in which the buckets are contiguous and overlap. However, many systems retrieve data from secondary storage in same-size blocks called pages. Fetching a page is a relatively expensive process; but once a page is fetched, its contents can be accessed orders of magnitude faster. We utilize this property of memory retrieval, presenting a variant of cuckoo hashing incorporating the following constraint: each bucket must be fully contained in a single page, but buckets are not necessarily contiguous. Empirical results show that this modification increases memory utilization and decreases the number of iterations required to insert an item. If each item is hashed to two buckets of capacity two, the page size is 8, and each bucket is fully contained in a single page, the memory utilization equals 89.71% in the classical contiguous disjoint bucket variant, 93.78% in the contiguous overlapping bucket variant, and increases to 97.46% in our new non-contiguous bucket variant. When the memory utilization is 92% and we use breadth first search to look for a vacant position, the number of iterations required to insert a new item is dramatically reduced from 545 in the contiguous overlapping buckets variant to 52 in our new non-contiguous bucket variant. In addition to the empirical results, we present a theoretical lower bound on the memory utilization of our variation as a function of the page size.

cs.DS↗

arXiv2011

Pattern Matching under Polynomial Transformation

We consider a class of pattern matching problems where a normalising transformation is applied at every alignment. Normalised pattern matching plays a key role in fields as diverse as image processing and musical information processing where application specific transformations are often applied to the input. By considering the class of polynomial transformations of the input, we provide fast algorithms and the first lower bounds for both new and old problems. Given a pattern of length m and a longer text of length n where both are assumed to contain integer values only, we first show O(n log m) time algorithms for pattern matching under linear transformations even when wildcard symbols can occur in the input. We then show how to extend the technique to polynomial transformations of arbitrary degree. Next we consider the problem of finding the minimum Hamming distance under polynomial transformation. We show that, for any epsilon>0, there cannot exist an O(n m^(1-epsilon)) time algorithm for additive and linear transformations conditional on the hardness of the classic 3SUM problem. Finally, we consider a version of the Hamming distance problem under additive transformations with a bound k on the maximum distance that need be reported. We give a deterministic O(nk log k) time solution which we then improve by careful use of randomisation to O(n sqrt(k log k) log n) time for sufficiently small k. Our randomised solution outputs the correct answer at every position with high probability.

cs.DS↗

arXiv2011

Sublinear Time, Measurement-Optimal, Sparse Recovery For All

An approximate sparse recovery system in ell_1 norm formally consists of parameters N, k, epsilon an m-by-N measurement matrix, Phi, and a decoding algorithm, D. Given a vector, x, where x_k denotes the optimal k-term approximation to x, the system approximates x by hat_x = D(Phi.x), which must satisfy ||hat_x - x||_1 <= (1+epsilon)||x - x_k||_1. Among the goals in designing such systems are minimizing m and the runtime of D. We consider the "forall" model, in which a single matrix Phi is used for all signals x. All previous algorithms that use the optimal number m=O(k log(N/k)) of measurements require superlinear time Omega(N log(N/k)). In this paper, we give the first algorithm for this problem that uses the optimum number of measurements (up to a constant factor) and runs in sublinear time o(N) when k=o(N), assuming access to a data structure requiring space and preprocessing O(N).

cs.DS↗

arXiv2011

Space Lower Bounds for Online Pattern Matching

We present space lower bounds for online pattern matching under a number of different distance measures. Given a pattern of length m and a text that arrives one character at a time, the online pattern matching problem is to report the distance between the pattern and a sliding window of the text as soon as the new character arrives. We require that the correct answer is given at each position with constant probability. We give Omega(m) bit space lower bounds for L_1, L_2, L_\infty, Hamming, edit and swap distances as well as for any algorithm that computes the cross-correlation/convolution. We then show a dichotomy between distance functions that have wildcard-like properties and those that do not. In the former case which includes, as an example, pattern matching with character classes, we give Omega(m) bit space lower bounds. For other distance functions, we show that there exist space bounds of Omega(log m) and O(log^2 m) bits. Finally we discuss space lower bounds for non-binary inputs and show how in some cases they can be improved.

cs.DS↗

arXiv2011

Even Better Framework for min-wise Based Algorithms

In a recent paper from SODA11 \cite{kminwise} the authors introduced a general framework for exponential time improvement of \minwise based algorithms by defining and constructing almost \kmin independent family of hash functions. Here we take it a step forward and reduce the space and the independent needed for representing the functions, by defining and constructing a \dkmin independent family of hash functions. Surprisingly, for most cases only 8-wise independent is needed for exponential time and space improvement. Moreover, we bypass the $O(\log{\frac{1}ε})$ independent lower bound for approximately \minwise functions \cite{patrascu10kwise-lb}, as we use alternative definition. In addition, as the independent's degree is a small constant it can be implemented efficiently. Informally, under this definition, all subsets of size $d$ of any fixed set $X$ have an equal probability to have hash values among the minimal $k$ values in $X$, where the probability is over the random choice of hash function from the family. This property measures the randomness of the family, as choosing a truly random function, obviously, satisfies the definition for $d=k=|X|$. We define and give an efficient time and space construction of approximately \dkmin independent family of hash functions. The degree of independent required is optimal, i.e. only $O(d)$ for $2 \le d < k=O(\frac{d}{ε^2})$, where $ε\in (0,1)$ is the desired error bound. This construction can be used to improve many \minwise based algorithms, such as \cite{sizeEstimationFramework,Datar02estimatingrarity,NearDuplicate,SimilaritySearch,DBLP:conf/podc/CohenK07}, as will be discussed here. To our knowledge such definitions, for hash functions, were never studied and no construction was given before.

cs.DS↗

arXiv2010

Fast Pseudo-Random Fingerprints

We propose a method to exponentially speed up computation of various fingerprints, such as the ones used to compute similarity and rarity in massive data sets. Rather then maintaining the full stream of $b$ items of a universe $[u]$, such methods only maintain a concise fingerprint of the stream, and perform computations using the fingerprints. The computations are done approximately, and the required fingerprint size $k$ depends on the desired accuracy $ε$ and confidence $δ$. Our technique maintains a single bit per hash function, rather than a single integer, thus requiring a fingerprint of length $k = O(\frac{\ln \frac{1}δ}{ε^2})$ bits, rather than $O(\log u \cdot \frac{\ln \frac{1}δ}{ε^2})$ bits required by previous approaches. The main advantage of the fingerprints we propose is that rather than computing the fingerprint of a stream of $b$ items in time of $O(b \cdot k)$, we can compute it in time $O(b \log k)$. Thus this allows an exponential speedup for the fingerprint construction, or alternatively allows achieving a much higher accuracy while preserving computation time. Our methods rely on a specific family of pseudo-random hashes for which we can quickly locate hashes resulting in small values.

cs.DS↗

arXiv2010

Fast Moment Estimation in Data Streams in Optimal Space

We give a space-optimal algorithm with update time O(log^2(1/eps)loglog(1/eps)) for (1+eps)-approximating the pth frequency moment, 0 < p < 2, of a length-n vector updated in a data stream. This provides a nearly exponential improvement in the update time complexity over the previous space-optimal algorithm of [Kane-Nelson-Woodruff, SODA 2010], which had update time Omega(1/eps^2).

cs.DS↗

arXiv2010

On the hardness of distance oracle for sparse graph

In this paper we show that set-intersection is harder than distance oracle on sparse graphs. Given a collection of total size n which consists of m sets drawn from universe U, the set-intersection problem is to build a data structure which can answer whether two sets have any intersection. A distance oracle is a data structure which can answer distance queries on a given graph. We show that if one can build distance oracle for sparse graph G=(V,E), which requires s(|V|,|E|) space and answers a (2-ε,c)-approximate distance query in time t(|V|,|E|) where (2-ε) is a multiplicative error and c is a constant additive error, then, set-intersection can be solved in t(m+|U|,n) time using s(m+|U|,n) space.

cs.DS↗

arXiv2010

Fast Set Intersection and Two Patterns Matching

In this paper we present a new problem, the fast set intersection problem, which is to preprocess a collection of sets in order to efficiently report the intersection of any two sets in the collection. In addition we suggest new solutions for the two-dimensional substring indexing problem and the document listing problem for two patterns by reduction to the fast set intersection problem.

cs.DS↗

arXiv2010

Range Non-Overlapping Indexing

We study the non-overlapping indexing problem: Given a text T, preprocess it so that you can answer queries of the form: given a pattern P, report the maximal set of non-overlapping occurrences of P in T. A generalization of this problem is the range non-overlapping indexing where in addition we are given two indexes i,j to report the maximal set of non-overlapping occurrences between these two indexes. We suggest new solutions for these problems. For the non-overlapping problem our solution uses O(n) space with query time of O(m + occ_{NO}). For the range non-overlapping problem we propose a solution with O(n\log^εn) space for some 0<ε<1 and O(m + \log\log n + occ_{ij,NO}) query time.

cs.DS↗

arXiv2009

Approximate Sparse Recovery: Optimizing Time and Measurements

An approximate sparse recovery system consists of parameters $k,N$, an $m$-by-$N$ measurement matrix, $Φ$, and a decoding algorithm, $\mathcal{D}$. Given a vector, $x$, the system approximates $x$ by $\widehat x =\mathcal{D}(Φx)$, which must satisfy $\| \widehat x - x\|_2\le C \|x - x_k\|_2$, where $x_k$ denotes the optimal $k$-term approximation to $x$. For each vector $x$, the system must succeed with probability at least 3/4. Among the goals in designing such systems are minimizing the number $m$ of measurements and the runtime of the decoding algorithm, $\mathcal{D}$. In this paper, we give a system with $m=O(k \log(N/k))$ measurements--matching a lower bound, up to a constant factor--and decoding time $O(k\log^c N)$, matching a lower bound up to $\log(N)$ factors. We also consider the encode time (i.e., the time to multiply $Φ$ by $x$), the time to update measurements (i.e., the time to multiply $Φ$ by a 1-sparse $x$), and the robustness and stability of the algorithm (adding noise before and after the measurements). Our encode and update times are optimal up to $\log(N)$ factors.

cs.DS↗

arXiv2008

Explicit Non-Adaptive Combinatorial Group Testing Schemes

Group testing is a long studied problem in combinatorics: A small set of $r$ ill people should be identified out of the whole ($n$ people) by using only queries (tests) of the form "Does set X contain an ill human?". In this paper we provide an explicit construction of a testing scheme which is better (smaller) than any known explicit construction. This scheme has $\bigT{\min[r^2 \ln n,n]}$ tests which is as many as the best non-explicit schemes have. In our construction we use a fact that may have a value by its own right: Linear error-correction codes with parameters $[m,k,δm]_q$ meeting the Gilbert-Varshamov bound may be constructed quite efficiently, in $\bigT{q^km}$ time.

cs.DS↗

arXiv2008

An Optimal Bloom Filter Replacement Based on Matrix Solving

We suggest a method for holding a dictionary data structure, which maps keys to values, in the spirit of Bloom Filters. The space requirements of the dictionary we suggest are much smaller than those of a hashtable. We allow storing n keys, each mapped to value which is a string of k bits. Our suggested method requires nk + o(n) bits space to store the dictionary, and O(n) time to produce the data structure, and allows answering a membership query in O(1) memory probes. The dictionary size does not depend on the size of the keys. However, reducing the space requirements of the data structure comes at a certain cost. Our dictionary has a small probability of a one sided error. When attempting to obtain the value for a key that is stored in the dictionary we always get the correct answer. However, when testing for membership of an element that is not stored in the dictionary, we may get an incorrect answer, and when requesting the value of such an element we may get a certain random value. Our method is based on solving equations in GF(2^k) and using several hash functions. Another significant advantage of our suggested method is that we do not require using sophisticated hash functions. We only require pairwise independent hash functions. We also suggest a data structure that requires only nk bits space, has O(n2) preprocessing time, and has a O(log n) query time. However, this data structures requires a uniform hash functions. In order replace a Bloom Filter of n elements with an error proability of 2^{-k}, we require nk + o(n) memory bits, O(1) query time, O(n) preprocessing time, and only pairwise independent hash function. Even the most advanced previously known Bloom Filter would require nk+O(n) space, and a uniform hash functions, so our method is significantly less space consuming especially when k is small.

cs.DS↗

arXiv2008

Approximating General Metric Distances Between a Pattern and a Text

Let $T=t_0 ... t_{n-1}$ be a text and $P = p_0 ... p_{m-1}$ a pattern taken from some finite alphabet set $Σ$, and let $\dist$ be a metric on $Σ$. We consider the problem of calculating the sum of distances between the symbols of $P$ and the symbols of substrings of $T$ of length $m$ for all possible offsets. We present an $ε$-approximation algorithm for this problem which runs in time $O(\frac{1}{ε^2}n\cdot \mathrm{polylog}(n,\absΣ))$

cs.DS↗

arXiv2008

Improved Deterministic Length Reduction

This paper presents a new technique for deterministic length reduction. This technique improves the running time of the algorithm presented in \cite{LR07} for performing fast convolution in sparse data. While the regular fast convolution of vectors $V_1,V_2$ whose sizes are $N_1,N_2$ respectively, takes $O(N_1 \log N_2)$ using FFT, using the new technique for length reduction, the algorithm proposed in \cite{LR07} performs the convolution in $O(n_1 \log^3 n_1)$, where $n_1$ is the number of non-zero values in $V_1$. The algorithm assumes that $V_1$ is given in advance, and $V_2$ is given in running time. The novel technique presented in this paper improves the convolution time to $O(n_1 \log^2 n_1)$ {\sl deterministically}, which equals the best running time given achieved by a {\sl randomized} algorithm. The preprocessing time of the new technique remains the same as the preprocessing time of \cite{LR07}, which is $O(n_1^2)$. This assumes and deals the case where $N_1$ is polynomial in $n_1$. In the case where $N_1$ is exponential in $n_1$, a reduction to a polynomial case can be used. In this paper we also improve the preprocessing time of this reduction from $O(n_1^4)$ to $O(n_1^3{\rm polylog}(n_1))$.

cs.DS↗