LSTM: A Search Space Odyssey

摘要

Several variants of the Long Short-Term Memory (LSTM) architecture forrecurrent neural networks have been proposed since its inception in 1995. Inrecent years, these networks have become the state-of-the-art models for avariety of machine learning problems. This has led to a renewed interest inunderstanding the role and utility of various computational components oftypical LSTM variants. In this paper, we present the first large-scale analysisof eight LSTM variants on three representative tasks: speech recognition,handwriting recognition, and polyphonic music modeling. The hyperparameters ofall LSTM variants for each task were optimized separately using random search,and their importance was assessed using the powerful fANOVA framework. Intotal, we summarize the results of 5400 experimental runs ($\approx 15$ yearsof CPU time), which makes our study the largest of its kind on LSTM networks.Our results show that none of the variants can improve upon the standard LSTMarchitecture significantly, and demonstrate the forget gate and the outputactivation function to be its most critical components. We further observe thatthe studied hyperparameters are virtually independent and derive guidelines fortheir efficient adjustment.