Demystifying Information-Theoretic Clustering

Greg Ver Steeg, Aram Galstyan, Fei Sha, Simon DeDeo
Proceedings of the 31st International Conference on Machine Learning, PMLR 32(1):19-27, 2014.

Abstract

We propose a novel method for clustering data which is grounded in information-theoretic principles and requires no parametric assumptions. Previous attempts to use information theory to define clusters in an assumption-free way are based on maximizing mutual information between data and cluster labels. We demonstrate that this intuition suffers from a fundamental conceptual flaw that causes clustering performance to deteriorate as the amount of data increases. Instead, we return to the axiomatic foundations of information theory to define a meaningful clustering measure based on the notion of consistency under coarse-graining for finite data.

Cite this Paper


BibTeX
@InProceedings{pmlr-v32-steeg14, title = {Demystifying Information-Theoretic Clustering}, author = {Ver Steeg, Greg and Galstyan, Aram and Sha, Fei and DeDeo, Simon}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {19--27}, year = {2014}, editor = {Xing, Eric P. and Jebara, Tony}, volume = {32}, number = {1}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/steeg14.pdf}, url = {https://proceedings.mlr.press/v32/steeg14.html}, abstract = {We propose a novel method for clustering data which is grounded in information-theoretic principles and requires no parametric assumptions. Previous attempts to use information theory to define clusters in an assumption-free way are based on maximizing mutual information between data and cluster labels. We demonstrate that this intuition suffers from a fundamental conceptual flaw that causes clustering performance to deteriorate as the amount of data increases. Instead, we return to the axiomatic foundations of information theory to define a meaningful clustering measure based on the notion of consistency under coarse-graining for finite data.} }
Endnote
%0 Conference Paper %T Demystifying Information-Theoretic Clustering %A Greg Ver Steeg %A Aram Galstyan %A Fei Sha %A Simon DeDeo %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-steeg14 %I PMLR %P 19--27 %U https://proceedings.mlr.press/v32/steeg14.html %V 32 %N 1 %X We propose a novel method for clustering data which is grounded in information-theoretic principles and requires no parametric assumptions. Previous attempts to use information theory to define clusters in an assumption-free way are based on maximizing mutual information between data and cluster labels. We demonstrate that this intuition suffers from a fundamental conceptual flaw that causes clustering performance to deteriorate as the amount of data increases. Instead, we return to the axiomatic foundations of information theory to define a meaningful clustering measure based on the notion of consistency under coarse-graining for finite data.
RIS
TY - CPAPER TI - Demystifying Information-Theoretic Clustering AU - Greg Ver Steeg AU - Aram Galstyan AU - Fei Sha AU - Simon DeDeo BT - Proceedings of the 31st International Conference on Machine Learning DA - 2014/01/27 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-steeg14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 32 IS - 1 SP - 19 EP - 27 L1 - http://proceedings.mlr.press/v32/steeg14.pdf UR - https://proceedings.mlr.press/v32/steeg14.html AB - We propose a novel method for clustering data which is grounded in information-theoretic principles and requires no parametric assumptions. Previous attempts to use information theory to define clusters in an assumption-free way are based on maximizing mutual information between data and cluster labels. We demonstrate that this intuition suffers from a fundamental conceptual flaw that causes clustering performance to deteriorate as the amount of data increases. Instead, we return to the axiomatic foundations of information theory to define a meaningful clustering measure based on the notion of consistency under coarse-graining for finite data. ER -
APA
Ver Steeg, G., Galstyan, A., Sha, F. & DeDeo, S.. (2014). Demystifying Information-Theoretic Clustering. Proceedings of the 31st International Conference on Machine Learning, in Proceedings of Machine Learning Research 32(1):19-27 Available from https://proceedings.mlr.press/v32/steeg14.html.

Related Material