grid cells
- observed to correspond to location tracking in rats, monkeys
- namely, the function of dead reckoning
- hippocampus and entorhinal cortex
- think of the grid cells as a literal regular grid
- the phase space of the grid cells is toroidal: you can shift the grid up and down and return to the same grid configuration
- data can be encoded by having multiple grid cell "modules", each with their own periods \(\lambda_i\)
- my questions:
- We talk about the mapping between external states and grid cell states changing, even though the internal states do not. Where is that mapping stored? How does that mapping arise?
- The hippocampus is involved in both navigation and episodic memory. Is the input to episodic memory the same as the velocity inputs?
- There are probably unfavorable selections of \(\lambda_i\) that limit the capacity of the system. Is the hope that this will not happen on average for a random selection of \(\lambda_i\) 's? Or is there some type of pressure to choose co-prime periods?
- my question for vectorHASH:
- What are the organization of the hashes? Are similar inputs given similar hashes?
- Once an input is associated with a basin, do you have to stay there? Can you deepen the basin?