Design LLM architectures based on the cortical column hypothesis - that the brain uses a repeated canonical microcircuit for all types of computation. This project would create modular, column-like structures that can be combined for different cognitive tasks.
Architecture details:
- Design a 'universal cortical column' module with 6-layer structure
- Implement lateral inhibition and winner-take-all dynamics
- Create hierarchical connections between columns
- Test modularity by swapping columns between tasks
- Compare with cortical recordings during language tasks
Applies fundamental neuroscience principles to AI architecture design.
Design LLM architectures based on the cortical column hypothesis - that the brain uses a repeated canonical microcircuit for all types of computation. This project would create modular, column-like structures that can be combined for different cognitive tasks.
Architecture details:
Applies fundamental neuroscience principles to AI architecture design.