The Explicit Decomposition with Neighborhoods (EDeN) is a decompositional kernel based on the Neighborhood Subgraph Pairwise Distance Kernel (NSPDK) that can be used to induce an explicit feature representation for graphs. This in turn allows the adoption of machine learning algorithm to perform supervised and unsupervised learning task in a scalable way (e.g. fast stochastic gradient descent methods in classification).
One of the novelties introduced in EDeN is the ability to take in input approximate kernel feature maps schemes to process nodes containing arbitrary data types. Another novelty is the ability to take in input weighted graphs.
Costa, Fabrizio, and Kurt De Grave. "Fast neighborhood subgraph pairwise distance kernel." Proceedings of the 26th International Conference on Machine Learning. 2010. (PDF)
P. Frasconi, F. Costa, K. De Grave, L. De Raedt,"kLog: A Language for Logical and Relational Learning with Kernels", Artificial Intelligence, 2014. (PDF)
A few examples can be found as IPython Notebook inside of the examples folder. For example see here:
explicit graph modelling
annotation
http://nbviewer.ipython.org/github/fabriziocosta/pyEDeN/blob/master/examples/annotation.ipynb
classification
http://nbviewer.ipython.org/github/fabriziocosta/pyEDeN/blob/master/examples/classification.ipynb
nearest neighbour and Gram matrix
learning curve
http://nbviewer.ipython.org/github/fabriziocosta/pyEDeN/blob/master/examples/learning_curve.ipynb
parameter exploration
serialization
http://nbviewer.ipython.org/github/fabriziocosta/pyEDeN/blob/master/examples/serialization.ipynb
string
http://nbviewer.ipython.org/github/fabriziocosta/pyEDeN/blob/master/examples/strings.ipynb