The unreasonable effectiveness of the forget gate
Neural and Evolutionary Computing
2018-09-14 v3 Machine Learning
Machine Learning
Abstract
Given the success of the gated recurrent unit, a natural question is whether all the gates of the long short-term memory (LSTM) network are necessary. Previous research has shown that the forget gate is one of the most important gates in the LSTM. Here we show that a forget-gate-only version of the LSTM with chrono-initialized biases, not only provides computational savings but outperforms the standard LSTM on multiple benchmark datasets and competes with some of the best contemporary models. Our proposed network, the JANET, achieves accuracies of 99% and 92.5% on the MNIST and pMNIST datasets, outperforming the standard LSTM which yields accuracies of 98.5% and 91%.
Keywords
Cite
@article{arxiv.1804.04849,
title = {The unreasonable effectiveness of the forget gate},
author = {Jos van der Westhuizen and Joan Lasenby},
journal= {arXiv preprint arXiv:1804.04849},
year = {2018}
}
Comments
Corrected LSTM gradient derivations. Added link to code