Orthogonal residual connections for long-term memory retention in recurrent neural networks
Andrea Ceni ⋅ Claudio Gallicchio
Abstract
Recurrent neural networks (RNNs) face difficulties learning long-term dependencies from sequential data and keeping working memory for long time horizons.Here we present an approach based on mathematically grounded principles to overcome this challenge via orthogonal residual connections.
Chat is not available.
Successful Page Load