Related Experiment Videos
Self-organization of associative memory and pattern classification: recurrent signal processing on topological
P Tavan1, H Grubmüller, H Kühnel
1Physik-Department, Technische Universität München, Garching, Federal Republic of Germany.
Biological Cybernetics
|January 1, 1990
Summary
This study introduces a novel neural network dynamics that transforms topological feature maps into auto-associative memories. This enables self-organized, hierarchical pattern classification and cluster analysis for complex data.
Area of Science:
- Computational Neuroscience
- Machine Learning
- Artificial Intelligence
Background:
- Topological maps store statistical data's probability densities.
- Extracting this information traditionally requires complex methods.
Purpose of the Study:
- To extend neural topological feature maps.
- To enable self-organization of auto-associative memory.
- To achieve hierarchical pattern classification.
Main Methods:
- Introduced a recurrent dynamics of signal processing.
- Applied this dynamics to topological maps.
- Developed a generalized neural network scheme.
Main Results:
- The dynamics convert topological maps into auto-associative memories.
- The memory performs cluster analysis on real-valued feature vectors.
- This represents a generalization of non-linear matrix-type associative memories.
Conclusions:
- The developed scheme enables a feature atlas concept.
- Facilitates self-organized, hierarchical pattern classification.
- Offers a novel approach to data analysis and memory