Improving Neural Language Models with a Continuous Cache
2016-12-13Code Available0· sign in to hype
Edouard Grave, Armand Joulin, Nicolas Usunier
Code Available — Be the first to reproduce this paper.
ReproduceCode
- github.com/Asteur/RERITES-AvgWeightDescentLSTM-PoetryGenerationpytorch★ 0
- github.com/llppff/ptb-lstmorqrnn-pytorchpytorch★ 0
- github.com/uclanlp/NamedEntityLanguageModelpytorch★ 0
- github.com/AtheMathmo/lookahead-lstmpytorch★ 0
- github.com/jb33k/awd-lstm-lm-ThinkNetpytorch★ 0
- github.com/SachinIchake/KALMpytorch★ 0
- github.com/jhave/RERITES-AvgWeightDescentLSTM-PoetryGenerationpytorch★ 0
- github.com/philippwirth/treelangrnnpytorch★ 0
- github.com/arvieFrydenlund/awd-lstm-lmpytorch★ 0
- github.com/ari-holtzman/genlmpytorch★ 0
Abstract
We propose an extension to neural network language models to adapt their prediction to the recent history. Our model is a simplified version of memory augmented networks, which stores past hidden activations as memory and accesses them through a dot product with the current hidden activation. This mechanism is very efficient and scales to very large memory sizes. We also draw a link between the use of external memory in neural network and cache models used with count based language models. We demonstrate on several language model datasets that our approach performs significantly better than recent memory augmented networks.
Tasks
Benchmark Results
| Dataset | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| WikiText-103 | Neural cache model (size = 2,000) | Test perplexity | 40.8 | — | Unverified |
| WikiText-103 | Neural cache model (size = 100) | Test perplexity | 44.8 | — | Unverified |
| WikiText-2 | Grave et al. (2016) - LSTM + continuous cache pointer | Test perplexity | 68.9 | — | Unverified |
| WikiText-2 | Grave et al. (2016) - LSTM | Test perplexity | 99.3 | — | Unverified |