Neural Information Processing

Models of Neural Development

opographic projections between neural sheets, orientation columns and ocular dominance columns in early visual areas have served as paradigmatic model systems for understanding the mechanisms underlying neural plasticity and development. Using mathematical models and computer simulations we investigated how activity driven and intrinsic processes interact in order to generate the observed anatomical connectivity patterns and response properties of neurons. We describe the development of those patterns as a goal-oriented (in the sense of underlying cost-functions) self-organizing process, which extracts information from the environment and imprints this knowledge into neural circuits. Particular emphasis was given to competitive networks including the Self-Organizing Map, which are known to trade smoothness vs. completeness of representations and which lead to patterns which fit experimental data surprisingly well.The mathematical properties of self-organizing maps were also analysed in a machine learning context. For details see "Research" page "Learning Vector Quantization and Self-organizing Maps"

Acknowledgements: Research was funded by BMBF, DFG, and the Technische Universität Berlin.