Neuroscience based machine intelligence models by Gary Gaulin

Monday, March 17, 2014


Grid Cell Attractor Network for place avoidance spatial navigation

This is a demonstration program (including Windows .exe) for the attractor network that made it possible for the next Intelligence Design Lab critter to confidently challenge the invisible moving shock zone arena, which required adding this hippocampus related network to its confidence system. It's the internal world model where the path around obstacles is planned out, visualized. With this added it learns to leave food in time to get out of the way of an approaching shock zone, then is soon impatiently waiting behind it until safe to eat the rest. Grid fields (here one field per cell) form a hexagonal array/lattice with an electrochemical field that is disrupted by border (also called boundary) cells mixed into the grid cell population to make places in the grid Repel, instead of Attract. As in radio transmission a cyan colored attracting location (food or other immediate need) emits continuous AC waves by turning off then on again, oscillates. Waves will propagate around the barrier to the yellow with tan inside critter location, which then guide it every step of the way back to the oscillating attractor by following the directional violet color angular vectors formed by activation pattern from the 6 neighboring cells around each place, along the way. How our brain or other cognitive system might produce and combine signals into such a spatial representation does not matter to this minimal code model. Since this demonstration greatly simplifies what is most important to know in regards to how the upcoming IDLab4 works it made sense to start with this simplified model, now ready for you to experiment with.



The primary Code:








No comments: