Simplified Long Short-term Memory Recurrent Neural Networks: part II
This work offers incremental improvements for researchers and practitioners in neural networks by reducing computational costs in LSTMs.
The authors introduced five simplified LSTM variants with reduced parameters, showing they perform comparably to standard LSTM on MNIST while using fewer parameters, and noting that these variants maintain accuracy with ReLU activation where standard LSTM drops after many epochs.
This is part II of three-part work. Here, we present a second set of inter-related five variants of simplified Long Short-term Memory (LSTM) recurrent neural networks by further reducing adaptive parameters. Two of these models have been introduced in part I of this work. We evaluate and verify our model variants on the benchmark MNIST dataset and assert that these models are comparable to the base LSTM model while use progressively less number of parameters. Moreover, we observe that in case of using the ReLU activation, the test accuracy performance of the standard LSTM will drop after a number of epochs when learning parameter become larger. However all of the new model variants sustain their performance.