Seeing is Believing (Parts 1 & 2)

The following topic from @afiser was recorded between two meetings:

Numenta Research Meeting Feb 19, 2020

Numenta Research Meeting Feb 24, 2020

1 Like

Sorry I messed up the first video postings. I think I have this sorted out now.

1 Like

The earlier Grid Cells for Conceptual Spaces information in another research meeting topic first made me think of upper level 1D left/right and forward/reverse navigational control of insects and ourselves. Authors mention:

Activity across a population of grid cells is thought to precisely encode the animal’s position, direction, and speed as it navigates an environment (Moser et al., 2008).

For the ID Lab model to work the “Activity across a population of grid cells” in this part of the brain encodes a conceptual/imagined position, direction, and speed that depends on what is mapped into the 2D space. The animal’s (circuit delay time compensated) actual predicted angular direction and speed is compared to the conceptualized direction and speed. Amount of error between the two sets of readings is used to adjust confidence level in motor actions.

The Experience-dependent spatial expectations in mouse visual cortex paper was to me describing V1 related direction and speed predictions needed for actual readings, while the first introduced Conceptual Spaces paper was best describing the conceptualized (where it wants to go) readings from the “navigation network” where place in the map it needs to travel to becomes active and rest propagate (or not) its waves according to the experience learned properties of what is at that place. The interaction of these two spaces through the (David L. Heiserman described) guess and confidence based motor system produces “trial and error learning” that makes it come alive.

Since the ID Lab model uses actual speed and direction to update position in the environment and conceptual map I did not need to model V1 and associated areas. Behavior would not change by another way calculating the same actual readings.

The “Grid Cells for Conceptual Spaces” would be needed in the 2D space for picturing how to get to where it wants to be, but not necessarily exactly work the same for V1 area for predicted position of a fast moving ball not where eye’s see it but the position it should in the future be located.

With the topics of your latest research meetings this way making sense to me I felt I need to explain this to you. At least might help better explain how the ID Lab critter works.

1 Like