Learning of Neural Information Routing for Correspondence Finding

Authors:
Jan D. Bouecke;Jörg Lücke
Affiliations:
Institute of Neural Information Processing, Universität Ulm, Ulm, Germany 89081 and Institute for Neuroinformatics, Ruhr-Universität Bochum, Bochum, Germany 44801;FIAS, Goethe-Universität Frankfurt, Frankfurt am Main, Germany 60438 and Institute for Theoretical Physics, Goethe-Universität Frankfurt, Frankfurt am Main, Germany 60438
Venue:
ICANN '08 Proceedings of the 18th international conference on Artificial Neural Networks, Part II
Year:
2008

Citing 9
Cited 1

SEEMORE: combining color, shape, and texture histogramming in a neurally inspired approach to visual object recognition

Neural Computation
Distortion Invariant Object Recognition in the Dynamic Link Architecture

IEEE Transactions on Computers
Face Authentication Test on the BANCA Database

ICPR '04 Proceedings of the Pattern Recognition, 17th International Conference on (ICPR'04) Volume 4 - Volume 04
Maplets for correspondence-based object recognition

Neural Networks - 2004 Special issue: New developments in self-organizing systems
A self-organizing map of sigma-pi units

Neurocomputing
Rapid convergence to feature layer correspondences

Neural Computation
A parallel computation that assigns canonical object-based frames of reference

IJCAI'81 Proceedings of the 7th international joint conference on Artificial intelligence - Volume 2
Dynamics of cortical columns – self-organization of receptive fields

ICANN'05 Proceedings of the 15th international conference on Artificial Neural Networks: biological Inspirations - Volume Part I
Rapid correspondence finding in networks of cortical columns

ICANN'06 Proceedings of the 16th international conference on Artificial Neural Networks - Volume Part I

Learning of lateral connections for representational invariant recognition

ICANN'10 Proceedings of the 20th international conference on Artificial neural networks: Part III

Quantified Score

Hi-index	0.00

Visualization

Abstract

We present a neural network model that learns to find correspondences. The network uses control units that gate feature information from an input to a model layer of neural feature detectors. The control units communicate via a network of lateral connections to coordinate the gating of feature information such that information about spatial feature arrangements can be used for correspondence finding. Using synaptic plasticity to modify the connections amongst control units, we show that the network can learn to find the relevant neighborhood relationship of features in a given class of input patterns. In numerical simulations we show quantitative results on pairs of one-dimensional artificial inputs and preliminary results on two-dimensional natural images. In both cases the system gradually learns the structure of feature neighborhood relations and uses this information to gradually improve in the task of correspondence finding.