Towards Non-saturating Recurrent Units for Modelling Long-term Dependencies
This work addresses a fundamental challenge in sequence modeling for machine learning, offering an incremental improvement over existing architectures like LSTM.
The authors tackled the problem of vanishing gradients in recurrent neural networks for long-term dependencies by proposing a Non-saturating Recurrent Unit (NRU) that avoids saturating activations and gates. They demonstrated that NRU consistently performs among the top 2 models across synthetic and real-world tasks with and without long-term dependencies.
Modelling long-term dependencies is a challenge for recurrent neural networks. This is primarily due to the fact that gradients vanish during training, as the sequence length increases. Gradients can be attenuated by transition operators and are attenuated or dropped by activation functions. Canonical architectures like LSTM alleviate this issue by skipping information through a memory mechanism. We propose a new recurrent architecture (Non-saturating Recurrent Unit; NRU) that relies on a memory mechanism but forgoes both saturating activation functions and saturating gates, in order to further alleviate vanishing gradients. In a series of synthetic and real world tasks, we demonstrate that the proposed model is the only model that performs among the top 2 models across all tasks with and without long-term dependencies, when compared against a range of other architectures.