Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training

Open in new window