I guess the backbone of HTM is predictive modeling in an invariant representational way, which is further strengthened by integration of different sensory inputs. Since GANs are extremely efficient at capturing /generating probability distributions of a sample. Have you guys thought about using GANs to do the predictions (generate a probability of on/off cells in columns of an HTM)?
GANs are awesome. It’s hard to reconcile adversarial gradient descent with the learning mechanisms available to the brain though.
GANs show some similarities to Turing learning. I think you could view feedback and feedfoward signals in a competitive environment within the neuron with LTP LTD as being similar to Turing learning.
By observing the behavior of the system as well as the behaviors produced by the models, two sets of data samples are obtained. The classifiers are rewarded for discriminating between these two sets, that is, for correctly categorizing data samples as either genuine or counterfeit. Conversely, the models are rewarded for ‘tricking’ the classifiers into categorizing their data samples as genuine. - [1603.04904] Turing learning: a metric-free approach to inferring behavior and its application to swarms
You could see the nearby neurons providing input as if from predictive models, these connections would be rewarded if they can positively influence the neuron regards a particular input. At the same time the neuron itself will have its output connections rewarded if it makes correct predictions. You’d have to see erroneous signals as counterfeit, while correct signals would be genuine. The erroneous signals either from feedforward or feedback would be punished while those correct would be rewarded. A kind of consensus, from a minimum number of coincident incoming signals, would determine genuine.