Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling

Benchmark Model Rank Results
language-modelling-on-penn-treebank-wordInan et al. (2016) - Variational RHN#33Test perplexity: 66.0Validation perplexity: 68.1
language-modelling-on-wikitext-2Inan et al. (2016) - Variational LSTM (tied) (h=650) + augmented loss#34Test perplexity: 87.0Validation perplexity: 91.5
language-modelling-on-wikitext-2Inan et al. (2016) - Variational LSTM (tied) (h=650)#35Test perplexity: 87.7Validation perplexity: 92.3