Machine learning for non-metric proximity data

This blog provides literature, algorithms and data sets for the analysis of (indefinite) proximity data. In machine learning kernels are given as proximity data. But if the proximity measure is non-metric most kernel approaches are inaccurate or fail. This blog shows ways how to deal with these so called indefinite, non-positive or non-psd proximity data, providing links to literature and algorithms. The final objective is to provide - Probabilistic Models in Pseudo-Euclidean Spaces (ProMoS)

Thursday, 27 November 2014

Benchmark sets for indefinite proximity learning

At the page Benchmark I have provided a matlab file with 13 proximity benchmarks
at various complexity, scale and from different application domains. The page contains
also a description and references to the original work behind the datasets.

Saturday, 6 September 2014

journal publication on conformal prediction for proximity data

The paper Adaptive Conformal Semi-Supervised Vector Quantization for Dissimilarity Data
published at Pattern Recognition is now online. We propose a conformal classifier for
multi-class semi-supervised learning of proximity data.
Update: The article won the College best publication award of the University of Birmingham College of Engineering and Physical Sciences

Publication at ICANN 2014

Accepted paper at ICANN 2014 about Discriminative Fast Soft Competitive Learning
The approach extends relational Soft Competitive Learning (aka Relational Neural Gas) by a supervised learning scheme which is more reliable than former approaches. The method can
be used to get a (semi-) supervised clustering of (large) proximity matrices.

Wednesday, 18 June 2014

Embedding of potentially asymmetric proximity data (visualization)

Published a paper with Marc Strickert and Eyke Hüllermeier about

Correlation-based embedding of pairwise score data

The approach can be applied to embed potentially asymmetric proximity (score) data preserving

correlation between the data. It also shows multiple theoretical insights into the conceptual behavior of alternative approaches.

http://www.sciencedirect.com/science/article/pii/S0925231214003956

Published paper at Neurocomputing

Published a paper "Learning vector quantization for (dis-)similarities"

focusing on a framework for (non-metric) proximity based learning using prototypes.
http://www.sciencedirect.com/science/article/pii/S0925231213010941

Poster presentation about non-metric data analysis @MiWoCi/WSOM'14

I will have a poster presentation at MiWoCi / Workshop on Self-Organizing Maps 2014
http://www.global.hs-mittweida.de/~wsom2014
The poster will address different approaches to deal with indefinite / non-psd / non-positive
proximity data.

Saturday, 8 February 2014

Special session at ESANN 2014

Special session at ESANN 2014 about Learning of structured and non-standard data join us in Bruges (Belgium), 23-25 April 2014