This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Alon–Matias–Szegedy | |
|---|---|
| Name | Alon–Matias–Szegedy |
| Authors | Noga Alon; Yossi Matias; Mario Szegedy |
| First published | 1996 |
| Field | Algorithms; Data streams; Randomized algorithms |
| Related | Count–Min sketch; Morris counter; Flajolet–Martin algorithm; Frequency moments |
Alon–Matias–Szegedy is a randomized streaming algorithm introduced by Noga Alon, Yossi Matias, and Mario Szegedy that estimates frequency moments of data streams using sublinear space, pioneering work in streaming algorithms and sketching techniques. The method provided the first nontrivial lower and upper bounds for estimating the second and higher frequency moments, connecting to research by Donald Knuth, Rivest–Shamir–Adleman, and later work such as the Count–Min sketch and Flajolet–Martin algorithm, influencing theoretical developments at venues like the ACM Symposium on Theory of Computing and FOCS.
The AM-S work arose from challenges in processing massive streams encountered in contexts like Internet packet monitoring, telecommunications billing, and database query optimization, where storing the entire multiset is infeasible; it built on probabilistic techniques from Paul Erdős-influenced combinatorics, methods from Alfred Aho-style algorithm design, and prior streaming heuristics related to Philippe Flajolet and G. Nigel Martin. The authors sought to formalize the task of approximating frequency moments defined by researchers in statistics and information theory while leveraging tools common in research by Leslie Valiant, Robert Tarjan, and Michael Rabin.
The core problem defines a data stream as a sequence of elements drawn from a universe indexed 1..n, as studied in works by Robert Floyd and Donald Knuth, with the i-th frequency f_i denoting occurrences; the k-th frequency moment F_k = sum_i f_i^k was formalized in the AM-S paper building on notions used by John von Neumann and Norbert Wiener. The goal is to estimate F_k within relative error ε with high probability using space sublinear in n, a constraint echoing limits proved in reductions related to communication complexity and hardness results from László Lovász and Andrew Yao.
AM-S proposes a randomized estimator using pairwise-independent hash functions and linear sketches influenced by Tomás Feder and Sudipto Guha approaches, maintaining compact counters updated per stream element analogous to techniques in Graham Cormode later work. The algorithm picks random sign functions and computes linear projections of the frequency vector, inspired by concepts from Jacob Ziv and Abraham Lempel in coding theory, then combines multiple independent estimators via median-of-means as in analyses related to Alfredo Huberman and Leslie Valiant to reduce variance.
AM-S proved that for k=2 the estimator uses polylogarithmic space and yields (1±ε) approximations with constant failure probability, a result that contrasts with lower bounds developed in subsequent research by Sanjeev Arora and Moses Charikar; for k>2 they gave near-optimal upper bounds and matching lower bounds within polylogarithmic factors, tying into complexity separations explored by Ronald Rivest and Michael Sipser. Their variance and tail bounds employed moment inequalities familiar from work by Paul Lévy and Sergei Bernstein, and they established connections to communication complexity lower bounds previously studied by Eitan Kushilevitz and Noam Nisan.
Subsequent variants include the Count–Sketch of Charikar, Chen, and Farach-Colton, the Count–Min sketch by Cormode and Muthukrishnan, and space-efficient counters like the Morris counter; extensions adapted AM-S ideas to heavy-hitters detection in network measurement studied by Netscape-era researchers and to entropy estimation related to Claude Shannon's information measures. Work by Piotr Indyk, Eric Price, and Anup Rao generalized the hash-based sketching paradigm to l_p-norm estimation and compressive sensing links investigated by David Donoho and Emmanuel Candès, while empirical systems research at Google and Amazon Web Services incorporated sketching primitives derived from AM-S for telemetry and analytics.
The AM-S framework influenced a broad array of applications including network traffic analysis in projects at Bell Labs, database approximate query processing in systems by Michael Stonebraker, and large-scale log analytics at Yahoo! and Facebook, guiding algorithmic toolkits used in Hadoop-based ecosystems and Apache Spark adaptations. Theoretical impact includes spawning an entire research thread on streaming lower bounds and sketch complexity pursued at conferences like STOC and SODA, and shaping curricula in courses by Tim Roughgarden, Leslie Valiant, and Sanjeev Arora on randomized algorithms and sublinear computation.
Category:Streaming algorithms Category:Randomized algorithms Category:Data stream mining