 |
Computational Intelligence - Learning with Neural Methods on Structured Data |
|
Computational Intelligence - Learning with Neural Methods on Structured Data |
Main contributors:
Barbara Hammer, Alessio Micheli (Universita di Pisa), Marc Strickert, Alessandro Sperduti (Universita di Pisa),
Publications:
See publications on Barbara's or Marc's page.
Main idea:
Recursive data such as directed acyclic graphs can be processed naturally with a self-organizing mechanism enhanced with a recursive processing dynamics according to the data structure. Specific implementations of this general idea can be found in the literature: the Temporal Kohonen Map as proposed by Chappell and Taylor, the recursive SOM as proposed by Voegtlin, or the SOM for structured data as proposed by Hagenbuchner, Sperduti, and Tsoi. We have developed a general framework for these approaches such that a uniform formulation of the dynamics, transferring of training algorithms from the standard case, and theoretical investigations are easily possible.
We are currently performing several comparisons and experiments of the approaches. However, you can already have a look at nice animations of the dynamic of topology representing mechanisms for simple vectors.