IPL 2016 - Special session on Indefinite proximity learning

Today real life data are often not given as vectorial data, but by means of proximities (similarities or dissimilarities) only, calculated by an appropriate proximity function. The pairwise proximity matrix may be a symmetric positive semidefinite matrix - and hence a kernel matrix but can be much more generic.
Often the underlying data may not exist in a vector space and the proximity function violates metric properties, leading to indefinite, potentially asymmetric proximity matrices, which can not directly be processed by classical machine learning algorithms - like kernel machines.
We can find these settings in various domains like the analysis of text documents - using e.g. the compression distance, the comparison of biological sequence data - with domain specific alignment measures, shape retrieval problems in robotics using the inner distance, the representation of structured data like graphs or trees and many other applications.
The recent technological developments, also in the context of big data, allow the generation of very large data sets. If the data are represented by non-metric proximities the processing becomes particular challenging, because many classical mathematical models require metric properties.
Dedicated processing strategies for non-metric proximity data (indefinite proximities, non-positive kernels, dissimilarity data) are of wide interest and the subject of this special session.
We encourage submission of papers on novel methods for (pre-) processing of non-metric kernels, structured data or in the field of non-metric dissimilarity learning by means of computational intelligence and machine learning approaches, including but not limited to:
data analysis and pattern recognition approaches for (indefinite) proximity data, structured data and dissimilarity learning
clustering, classification, regression, embedding approaches for indefinite data
approaches in the line of matrix completion, collaborative filtering, reduction techniques for non-standard data
low rank matrix approaches - valid for indefinite proximities
vector space embedding
metric nearness and correction approaches
large scale proximity matrix analysis and handling
quality and error measures for indefinite data representations
applications with indefinite proximity data