Related Experiment Videos
Scaling self-organizing maps to model large cortical networks.
James A Bednar1, Amol Kelkar, Risto Miikkulainen
1Department of Computer Sciences, The University of Texas at Austin, Austin, TX 78712, USA. jbednar@inf.ed.ac.uk
Neuroinformatics
|September 15, 2004
Summary
This study introduces parameter scaling and a new growing map method (GLISSOM) to make large computational models of the visual cortex practical. These techniques enable detailed, large-scale simulations for studying complex visual phenomena.
Area of Science:
- Computational neuroscience
- Neuroscience
- Computer science
Background:
- Self-organizing computational models explain visual cortex features like topographic maps.
- Large-scale simulations of these models are computationally intensive, limiting studies of complex visual phenomena.
Purpose of the Study:
- Introduce techniques to make large-scale simulations of self-organizing models practical.
- Enable detailed studies of phenomena like object recognition and optic flow.
- Facilitate comparison of visual maps across species.
Main Methods:
- Derived parameter scaling equations for laterally connected self-organizing models.
- Developed a new growing map method called GLISSOM, reducing computational and memory requirements.
- Implemented these techniques in the Topographica simulator.
Main Results:
- Parameter scaling allows for quantitatively equivalent maps across different simulation sizes.
- GLISSOM significantly reduces resource needs, making whole human V1 simulation feasible on workstations.
- These methods facilitate debugging and scaling of complex neural models.
Conclusions:
- Parameter scaling and GLISSOM overcome computational barriers in modeling the visual cortex.
- These advancements pave the way for detailed, large-scale simulations of topographic maps.
- The Topographica simulator will aid research into complex visual processing and inter-species comparisons.