Papers
Topics
Authors
Recent
Search
2000 character limit reached

Greedy bi-criteria approximations for kk-medians and kk-means

Published 21 Jul 2016 in cs.DS and cs.LG | (1607.06203v1)

Abstract: This paper investigates the following natural greedy procedure for clustering in the bi-criterion setting: iteratively grow a set of centers, in each round adding the center from a candidate set that maximally decreases clustering cost. In the case of kk-medians and kk-means, the key results are as follows. ∙\bullet When the method considers all data points as candidate centers, then selecting O(klog⁡(1/ε))\mathcal{O}(k\log(1/\varepsilon)) centers achieves cost at most 2+ε2+\varepsilon times the optimal cost with kk centers. ∙\bullet Alternatively, the same guarantees hold if each round samples O(k/ε5)\mathcal{O}(k/\varepsilon^5) candidate centers proportionally to their cluster cost (as with kmeans++\texttt{kmeans++}, but holding centers fixed). ∙\bullet In the case of kk-means, considering an augmented set of n⌈1/ε⌉n^{\lceil1/\varepsilon\rceil} candidate centers gives 1+ε1+\varepsilon approximation with O(klog⁡(1/ε))\mathcal{O}(k\log(1/\varepsilon)) centers, the entire algorithm taking O(dklog⁡(1/ε)n1+⌈1/ε⌉)\mathcal{O}(dk\log(1/\varepsilon)n^{1+\lceil1/\varepsilon\rceil}) time, where nn is the number of data points in Rd\mathbb{R}^d. ∙\bullet In the case of Euclidean kk-medians, generating a candidate set via nO(1/ε2)n^{\mathcal{O}(1/\varepsilon^2)} executions of stochastic gradient descent with adaptively determined constraint sets will once again give approximation 1+ε1+\varepsilon with O(klog⁡(1/ε))\mathcal{O}(k\log(1/\varepsilon)) centers in dklog⁡(1/ε)nO(1/ε2)dk\log(1/\varepsilon)n^{\mathcal{O}(1/\varepsilon^2)} time. Ancillary results include: guarantees for cluster costs based on powers of metrics; a brief, favorable empirical evaluation against kmeans++\texttt{kmeans++}; data-dependent bounds allowing 1+ε1+\varepsilon in the first two bullets above, for example with kk-medians over finite metric spaces.

Authors (2)
Citations (8)

Summary

No one has generated a summary of this paper yet.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.