Academic and research sources
[S1] Neural Computation ↗
Hochreiter and Schmidhuber (1997) -- Long short-term memory. Introduced the gated cell architecture solving the vanishing gradient problem.
Neural Computation · 1997
[S2] Neural Computation ↗
Gers, Schmidhuber, and Cummins (2000) -- Learning to forget: continual prediction with LSTM. Added the forget gate, which was not in the original 1997 architecture.
Neural Computation · 2000
[S3] arXiv ↗
Graves (2013) -- Generating sequences with recurrent neural networks. Extended LSTMs with deep stacking and demonstrated sequence generation capabilities.
arXiv · 2013
[S4] Advances in Neural Information Processing Systems ↗
Sutskever, Vinyals, and Le (2014) -- Sequence to sequence learning with neural networks. Established the encoder-decoder LSTM architecture for sequence-to-sequence tasks.
Advances in Neural Information Processing Systems · 2014
[S5] Advances in Neural Information Processing Systems ↗
Gal and Ghahramani (2016) -- A theoretically grounded application of dropout in recurrent neural networks. Introduced variational dropout for LSTMs with the same mask at each time step.
Advances in Neural Information Processing Systems · 2016
[S6] International Journal of Forecasting ↗
Salinas, Flunkert, Gasthaus, and Januschowski (2020) -- DeepAR: probabilistic forecasting with autoregressive recurrent networks. LSTM-based probabilistic time-series forecasting at scale.
International Journal of Forecasting · 2020
[S7] PLOS ONE ↗
Makridakis, Spiliotis, and Assimakopoulos (2018) -- Statistical and machine learning forecasting methods: concerns and ways forward (PLOS ONE). M3-based comparison of ML methods, including recurrent networks, against statistical benchmarks; the M4 competition results appear in their separate 2018 IJF article.
PLOS ONE · 2018