Path-Normalized Optimization of Recurrent Neural Networks with ReLU Activations
Behnam Neyshabur · Yuhuai Wu · Russ Salakhutdinov · Nati Srebro
Keywords:
Deep Learning or Neural Networks
2016 Poster
Abstract
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.
Chat is not available.
Successful Page Load