A new tree kernel based on SOM-SD

Authors:
Fabio Aiolli;Giovanni Da San Martino;Alessandro Sperduti
Affiliations:
Dept. of Pure and Applied Mathematics, Padova, Italy;Dept. of Pure and Applied Mathematics, Padova, Italy;Dept. of Pure and Applied Mathematics, Padova, Italy
Venue:
ICANN'10 Proceedings of the 20th international conference on Artificial neural networks: Part II
Year:
2010

Citing 8
Cited 1

Support-Vector Networks

Machine Learning
Self-Organizing Maps

Self-Organizing Maps
A survey of kernels for structured data

ACM SIGKDD Explorations Newsletter
New ranking algorithms for parsing and tagging: kernels over discrete structures, and the voted perceptron

ACL '02 Proceedings of the 40th Annual Meeting on Association for Computational Linguistics
Report on the XML mining track at INEX 2005 and INEX 2006: categorization and clustering of XML documents

ACM SIGIR Forum
Route kernels for trees

ICML '09 Proceedings of the 26th Annual International Conference on Machine Learning
A self-organizing map for adaptive processing of structured data

IEEE Transactions on Neural Networks
Learning Nonsparse Kernels by Self-Organizing Maps for Structured Data

IEEE Transactions on Neural Networks

Tree Echo State Networks

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Many different paradigms have been studied in the past to treat tree structured data, including kernel and neural based approaches. However, both types of methods have their own drawbacks. Kernels typically can only cope with discrete labels and tend to be sparse. On the other side, SOM-SD, an extension of the SOM for structured data, is unsupervised and Markovian, i.e. the representation of a subtree does not consider where the subtree appears in a tree. In this paper, we present a hybrid approach which tries to overcome these problems. In particular, we propose a new kernel based on SOM-SD which adds information about the relative position of subtrees (the route) to the activation of the nodes in such a way to discriminate even those subtrees originally encoded by the same prototypes. Experiments have been performed against two well known benchmark datasets with promising results.