期刊名称：International Journal of Advances in Engineering and Management
电子版ISSN：2395-5252
出版年度：2020
卷号：2
期号：5
页码：618-622
DOI：10.35629/5252-0205548555
语种：English
出版社：IJAEM JOURNAL
摘要：Long-Term Memory (LSTM) is a specific neural network (RNN) organization designed to model temporary sequences and their long-distance dependencies more accurately than conventional RNNs. This paper, has examined the structures of the LSTM RNN and made some changes in its performance better. LSTM RNNs work better than DNNs. Here, some changes have been made to the gate count and remove some unnecessary elements for the standard LSTM design. This structure effectively uses model parameters than other imaginable, rapidly changing, and exudes an in-depth neural network feed with the order of the largest parameters. LSTM final loss is less than the final LSTM loss.
