My primary collaborator for the last six or so years has been my wife Crisi, who is engaging in her own project of similar scale. Our thoughts are so throroughly intertwined that I could not begin to list specifics.
A close friend since age eighteen, Eric Doehne, has always been a wonderful person to brainstorm with. Over the last few years he has also given me invaluable comments on various article drafts, en route to sculpting my ideas in their present form.
Kwabena Boahen has been a doubly helpful collaborator, as the only fellow neuroscientist/technologist whom I can ask anything and expect a really useful answer. His own eclectic interests are reflected in his research group, who have kept my slide-viewing and question-asking skills sharp.
Bruno Olshausen, also a friend since grad school, has been an invaluable colleague and interlocutor since we reunited at the Redwood Neuroscience Institute. My own interest in natural statistics and native 3-D processing follows his.
For a few years I paired up with Zhonghao (Kevin) Yang, as in our spare time we coded (in Python) a software reference implementation of the modular compression/prediction architecture I surmised was the secret to the brain. The simulation ultimately succeeded, and I presented results at Cal's Redwood Center, showing an autonomous system which "learned" a continuous-time, continuous-space 2-D manifold from a 64-dimensional (8x8) pixel array. It made simple motion-extrapolation predictions in the low-dimensional space, before uncompressing them back to predict actual input activity of the pixels themselves. It worked, but with such great difficulty that I realized no reasonable brain would ever force itself to learn from scratch that the world is in fact three-dimensional, when that assumption could have been built in for free. In my work with Kevin, as with my much earlier work trying to make LIF neurons fire irregularly, it was my persistent, deep-seated failures to implement a model I actually believed in which made me abandon it.