This excellent interview let us glimpse the exciting new developments in HTM theory! Thanks!
Part 6 (HTM on hardware) left me with a question however. The HICANN chip literature indicates that the hardware doesn’t directly implement “neurons”, but “membrane circuits”; one or many of those can be configured to act together as a single neuron and each membrane circuit implements an adaptive exponential firing model. My understanding is that their grouping implements an OR gate (although I’m not sure on this). Furthermore Sebastian’s paper shows how to connect membrane circuits together to implement minicolumns that exhibit the HTM theoretical behavior (including predictive states). I thought these behaviors cover the functionality of active dendrites.
What is meant, then, by “active dendrites” and by the current hardware not implementing them?
Should this question perhaps be moved to the thread with Sebastian’s paper?