Related Experiment Videos
Synaptic weight normalization effects for topographic mapping formation
1Kinki Polytechnic College Shiga, 1414, Hurukawa, Omihachiman, Shiga 523-8510, Japan. sakamoto@mac.com
Summary
We introduce a model for topographic mapping between cell layers using weight normalization. Output normalization enhances stability, allowing more learning rules to achieve topographic mappings compared to input normalization or no normalization.
Area of Science:
- Computational neuroscience
- Machine learning
Background:
- Topographic mapping preserves spatial relationships between neural layers.
- Weight normalization is crucial for stable neural network learning.
Purpose of the Study:
- To propose a flexible topographic mapping formation model.
- To investigate the impact of weight normalization on learning rules and topographic stability.
Main Methods:
- Developed a model with arbitrary neighborhood relations using undirected graphs.
- Defined learning rules and input/output weight normalization methods.
- Examined Hebbian and non-Hebbian weight modifications.
Main Results:
- Topographic mapping stability under correlational learning rules was observed with input normalization or no normalization.
- Output normalization enabled stability with both correlational and non-correlational learning rules.
- Computer simulations showed output normalization expands the set of learning rules yielding topographic mappings.
Conclusions:
- Weight normalization, particularly output normalization, significantly enhances the flexibility and stability of topographic mapping formation.
- The proposed model offers a versatile framework for understanding and implementing topographic maps in neural systems.