A Theoretically Grounded Application of Dropout in Recurrent Neural Networks

Benchmark Model Rank Results
language-modelling-on-penn-treebank-wordGal & Ghahramani (2016) - Variational LSTM (large)#34Test perplexity: 75.2Validation perplexity: 77.9
language-modelling-on-penn-treebank-wordGal & Ghahramani (2016) - Variational LSTM (medium)#37Test perplexity: 79.7Validation perplexity: 81.9