A NON-PARAMETRIC APPROACH TO SIMPLICITY CLUSTERING

Authors:
Peter Hines;Emmanuel M. Pothos;Nick Chater
Affiliations:
Department of Computer Science, University of York, York, U.K.;Department of Psychology, University of Wales, Swansea, U.K.;Department of Psychology, University College London, London, U.K.
Venue:
Applied Artificial Intelligence
Year:
2007

Citing 10
Cited 2

Inferring decision trees using the minimum description length principle

Information and Computation
Complexity optimized data clustering by competitive neural networks

Neural Computation
Clustering data by melting

Neural Computation
An introduction to Kolmogorov complexity and its applications (2nd ed.)

An introduction to Kolmogorov complexity and its applications (2nd ed.)
Clustering Algorithms

Clustering Algorithms
Knowledge Acquisition Via Incremental Conceptual Clustering

Machine Learning
On Prediction by Data Compression

ECML '97 Proceedings of the 9th European Conference on Machine Learning
Finding minimum entropy codes

Neural Computation
Iterative optimization and simplification of hierarchical clusterings

Journal of Artificial Intelligence Research
Minimum description length induction, Bayesianism, and Kolmogorov complexity

IEEE Transactions on Information Theory

The Simplicity and Power model for inductive inference

Artificial Intelligence Review
Occam and Bayes in predicting category intuitiveness

Artificial Intelligence Review

Quantified Score

Hi-index	0.00

Visualization

Abstract

The simplicity principle-an updating of Ockham's razor to take into account modern information theory-states that the preferred theory for a set of data is the one that allows for the most efficient encoding of the data. We consider this in the context of classification, or clustering, as a data reduction technique that helps describe a set of objects by dividing the objects into groups. The simplicity model we present favors clusters such that the similarity of the items in the clusters is maximal, while the similarity of items between clusters is minimal. Several novel features of our clustering criterion make it especially appropriate for clustering of data derived from, psychological procedures (e.g., similarity ratings): It is non-parametric, and may be applied in situations where the metric axioms are violated without requiring (information-forgetting) transformation procedures. We illustrate the use of the criterion with a selection of data sets. A distinctive aspect of this research is that it motivates a clustering algorithm from psychological principles.