This approach to ML seems very similar to HTM

[Cerebro]

They are modeling the neuron as predicting its own input in place of traditional back prop. I really enjoyed the talk.

[rhyolight]

I talked to Blake a bit about this in this podcast.