Capacity and Trainability in Recurrent Neural Networks
This work addresses the problem of understanding and improving RNN capacity and trainability for researchers and practitioners in machine learning, though it is incremental in refining existing knowledge.
The study experimentally demonstrates that common RNN architectures achieve similar capacity bounds, storing about 5 bits per parameter for task information and one real number per hidden unit for input history, with performance often limited by per-task parameter capacity rather than architectural differences. It also finds vanilla RNNs are harder to train but have slightly higher capacity, and proposes two new architectures, one easier to train than LSTM or GRU for deep stacks.
Two potential bottlenecks on the expressiveness of recurrent neural networks (RNNs) are their ability to store information about the task in their parameters, and to store information about the input history in their units. We show experimentally that all common RNN architectures achieve nearly the same per-task and per-unit capacity bounds with careful training, for a variety of tasks and stacking depths. They can store an amount of task information which is linear in the number of parameters, and is approximately 5 bits per parameter. They can additionally store approximately one real number from their input history per hidden unit. We further find that for several tasks it is the per-task parameter capacity bound that determines performance. These results suggest that many previous results comparing RNN architectures are driven primarily by differences in training effectiveness, rather than differences in capacity. Supporting this observation, we compare training difficulty for several architectures, and show that vanilla RNNs are far more difficult to train, yet have slightly higher capacity. Finally, we propose two novel RNN architectures, one of which is easier to train than the LSTM or GRU for deeply stacked architectures.