On Knowledge Distillation From Complex NetworksPublicationsTags: response prediction, SAM-mul Train, SAM-add Train, BIDAF, QANET