Online Facility Location on Semi-Random Streams

TL;DR

This paper analyzes online facility location in semi-random streams, establishing tight bounds on algorithm performance as stream order varies, and extends results to clustering problems with practical efficiency.

Contribution

It provides a tight analysis of the online facility location algorithm's performance in semi-random streams and develops efficient clustering algorithms adaptable to various dissimilarity measures.

Findings

Expected competitive ratio of $O(rac{ ext{log } t}{ ext{log log } t})$ for online facility location

Matching lower bound of $ ext{Omega}(rac{ ext{log } t}{ ext{log log } t})$ for any randomized algorithm

Optimal algorithms for random-order streams with $O(k)$ space and $O(nk)$ time

Abstract

In the streaming model, the order of the stream can significantly affect the difficulty of a problem. A $t$-semirandom stream was introduced as an interpolation between random-order ($t=1$) and adversarial-order ($t=n$) streams where an adversary intercepts a random-order stream and can delay up to $t$ elements at a time. IITK Sublinear Open Problem \#15 asks to find algorithms whose performance degrades smoothly as $t$ increases. We show that the celebrated online facility location algorithm achieves an expected competitive ratio of $O(\frac{\log t}{\log \log t})$. We present a matching lower bound that any randomized algorithm has an expected competitive ratio of $\Omega(\frac{\log t}{\log \log t})$. We use this result to construct an $O(1)$-approximate streaming algorithm for $k$-median clustering that stores $O(k \log t)$ points and has $O(k \log t)$ worst-case update time. Our technique generalizes to any dissimilarity measure that satisfies a weak triangle inequality, including $k$-means, $M$-estimators, and $\ell_p$ norms. The special case $t=1$ yields an optimal $O(k)$ space algorithm for random-order streams as well as an optimal $O(nk)$ time algorithm in the RAM model, closing a long line of research on this problem.