Path-Normalized Optimization of Recurrent Neural Networks with ReLU Activations

05/23/2016

∙

by   Behnam Neyshabur, et al.

∙

0

∙

share

We investigate the parameter-space geometry of recurrent neural networks (RNNs), and develop an adaptation of path-SGD optimization method, attuned to this geometry, that can learn plain RNNs with ReLU activations. On several datasets that require capturing long-term dependency structure, we show that path-SGD can significantly improve trainability of ReLU RNNs compared to RNNs trained with SGD, even with various recently suggested initialization schemes.