Faster Algorithms for the Constrained k-means Problem

Abstract

The classical center based clustering problems such as k-means/median/center assume that the optimal clusters satisfy the locality property that the points in the same cluster are close to each other. A number of clustering problems arise in machine learning where the optimal clusters do not follow such a locality property. For instance, consider the r -gather clustering problem where there is an additional constraint that each of the clusters should have at least r points or the capacitated clustering problem where there is an upper bound on the cluster sizes. Consider a variant of the k-means problem that may be regarded as a general version of such problems. Here, the optimal clusters O ₁, ..., O _k are an arbitrary partition of the dataset and the goal is to output k-centers c ₁, ..., c _k such that the objective function ${\sum }_{i = 1}^{k} {\sum }_{x \in O_{i}} ||x - c_{i}||^{2}$ is minimized. It is not difficult to argue that any algorithm (without knowing the optimal clusters) that outputs a single set of k centers, will not behave well as far as optimizing the above objective function is concerned. However, this does not rule out the existence of algorithms that output a list of such k centers such that at least one of these k centers behaves well. Given an error parameter ε > 0, let ℓ denote the size of the smallest list of k-centers such that at least one of the k-centers gives a (1 + ε) approximation w.r.t. the objective function above. In this paper, we show an upper bound on ℓ by giving a randomized algorithm that outputs a list of $2^{\tilde {O}(k/\varepsilon )}$ k-centers. We also give a closely matching lower bound of $2^{\tilde {\Omega }(k/\sqrt {\varepsilon })}$. Moreover, our algorithm runs in time $O \left (n d \cdot 2^{\tilde {O}(k/\varepsilon )} \right )$. This is a significant improvement over the previous result of Ding and Xu (2015) who gave an algorithm with running time O(n d ⋅ (log n)^k ⋅ 2^poly(k/ε)) and output a list of size O((log n)^k ⋅ 2^poly(k/ε)). Our techniques generalize for the k-median problem and for many other settings where non-Euclidean distance measures are involved.