Architectural Teacher Forcing

Similar to ECNN we use this concept to propagate the error from the previous state to the next states. However because we are using only one matrix in HCNNs the information gained through it is not lost for future predictions.