Single stream parallelization of generalized LSTM-like RNNs on a GPU
This work addresses the problem of slow RNN training for machine learning practitioners, offering an incremental improvement in parallelization techniques.
The paper tackles the challenge of long training times for RNNs by proposing a generalized graph-based structure covering LSTM networks and a parallelization approach that analyzes graph structures to exploit parallelism, achieving significant speed-ups with a single training stream and further acceleration with multiple streams.
Recurrent neural networks (RNNs) have shown outstanding performance on processing sequence data. However, they suffer from long training time, which demands parallel implementations of the training procedure. Parallelization of the training algorithms for RNNs are very challenging because internal recurrent paths form dependencies between two different time frames. In this paper, we first propose a generalized graph-based RNN structure that covers the most popular long short-term memory (LSTM) network. Then, we present a parallelization approach that automatically explores parallelisms of arbitrary RNNs by analyzing the graph structure. The experimental results show that the proposed approach shows great speed-up even with a single training stream, and further accelerates the training when combined with multiple parallel training streams.