Part of AGI experiments
Active Dendrites
Neuron models that fire more readily given the right top-down context - a mechanism for mixing prediction with signal.
Active-dendrite neuron models make a neuron more prone to fire when the appropriate context arrives at its dendrites. That gives a concrete mechanism for mixing top-down predictions with the bottom-up signal: predictions prime a set of neurons via dendritic activation, and the patterns those neurons are tuned to then drive network activity - effectively attending to the predicted parts of the input. The experiments explore how this produces sparse latent representations, with an eye toward the attention-like behavior we use to, say, follow one voice in a noisy room.