Simple and Deterministic Matrix Sketches

Lecture / Panel
For NYU Community

Speaker: Edo Liberty, Yahoo! Labs


A sketch of a matrix $A$ is another matrix $B$ which is significantly smaller than $A$, but still approximates it well. Finding such sketches efficiently is an important building block in modern algorithms for approximating, for example, the PCA of massive matrices. This task is made more challenging in the streaming model, where each row of the input matrix can be processed only once and storage is severely limited. In this talk, we adapt a well known streaming algorithm for approximating item frequencies to the matrix sketching setting. The algorithm receives $n$ rows of a large matrix $A$ in $R^{n times m}$ one after the other, in a streaming fashion. It maintains a sketch $B$ in $R^{ell times m}$ containing only $ell << n$ rows, but still guarantees that $A^T A$ approximates $B^T B$. More accurately,

for all x, ||x||=1:  0 <= ||Ax||^2 - ||Bx||^2 <= 2||A||_{f}^2/ell

This algorithm's error decays proportionally to $1/ell$ using $O(m ell)$ space. In comparison, random-projection, hashing or sampling based algorithms produce convergence bounds proportional to $1/sqrt{ell}$. Sketch updates per row in $A$ require amortized $O(m ell)$ operations and the algorithm is perfectly parallelizable. Our experiments corroborate the algorithm's scalability and improved convergence rate. The presented algorithm also stands out in that it is deterministic, simple to implement, and elementary to prove.


Edo received his B.Sc in Physics and Computer Science from Tel Aviv University and his Ph.D in Computer Science from Yale University, under the supervision of Steven Zucker. During his PhD he spent time at both UCLA and Google as a researcher and an engineer. After that, he joined the Program in Applied Mathematics at Yale as a Post-Doctoral fellow. In 2009 he joined Yahoo! Labs in Israel. He recently moved to New York to lead the scalable machine learning group which focuses on the theory and practice of (very) large scale data mining and machine learning, and in particular, on theoretical foundation of machine learning, optimizations, scalable scientific computing, and machine learning systems and platforms.