Fig 3. Three level HPC model network.

Each of nine level 1 PE modules sends prediction to level 0 by feedback connection and receives error signal corresponding to their local image patches by a feed-forward connection. On the other hand, the PE module at level 2 sends the prediction signal to all level 1 modules and in reply, receives the error signal collectively from all these modules. Level 2, therefore, encodes larger visual patch and hence possess the larger receptive field.