New method trains recurrent networks faster without slow sequential learning

Original: Pretraining Recurrent Networks without Recurrence

Source: arxiv.org ↗