On evolutionary approaches to unsupervised nearest neighbor regression

Authors:
Oliver Kramer
Affiliations:
Fakultät II, Department for Computer Science, Carl von Ossietzky Universität Oldenburg, Oldenburg, Germany
Venue:
EvoApplications'12 Proceedings of the 2012t European conference on Applications of Evolutionary Computation
Year:
2012

Citing 7
Cited 1

Evolution strategies –A comprehensive introduction

Natural Computing: an international journal
Regularized principal manifolds

The Journal of Machine Learning Research
Principal Surfaces from Unsupervised Kernel Regression

IEEE Transactions on Pattern Analysis and Machine Intelligence
Probabilistic Non-linear Principal Component Analysis with Gaussian Process Latent Variable Models

The Journal of Machine Learning Research
Variants of unsupervised kernel regression: General cost functions

Neurocomputing
A derandomized approach to self-adaptation of evolution strategies

Evolutionary Computation
Dimensionality Reduction by Unsupervised K-Nearest Neighbor Regression

ICMLA '11 Proceedings of the 2011 10th International Conference on Machine Learning and Applications and Workshops - Volume 01

Unsupervised nearest neighbors with kernels

KI'12 Proceedings of the 35th Annual German conference on Advances in Artificial Intelligence

Quantified Score

Hi-index	0.00

Visualization

Abstract

The detection of structures in high-dimensional data has an important part to play in machine learning. Recently, we proposed a fast iterative strategy for non-linear dimensionality reduction based on the unsupervised formulation of K-nearest neighbor regression. As the unsupervised nearest neighbor (UNN) optimization problem does not allow the computation of derivatives, the employment of direct search methods is reasonable. In this paper we introduce evolutionary optimization approaches for learning UNN embeddings. Two continuous variants are based on the CMA-ES employing regularization with domain restriction, and penalizing extension in latent space. A combinatorial variant is based on embedding the latent variables on a grid, and performing stochastic swaps. We compare the results on artificial dimensionality reduction problems.