Improved Language Modeling by Decoding the Past

doi:doi:10.57702/fskxprc7

Improved Language Modeling by Decoding the Past

Highly regularized LSTMs achieve impressive results on several benchmark datasets in language modeling. We propose a new regularization method based on decoding the last token in the context using the predicted distribution of the next token.

Data and Resources

Original MetadataJSON
The json representation of the dataset with its distributions based on DCAT.
Explore
- Preview
- Download

Cite this as

Siddhartha Brahma (2024). Dataset: Improved Language Modeling by Decoding the Past. https://doi.org/10.57702/fskxprc7

DOI retrieved: December 16, 2024

Additional Info

Field	Value
Created	December 16, 2024
Last update	December 16, 2024
Defined In	https://doi.org/10.48550/arXiv.1808.05908
Author	Siddhartha Brahma
Homepage	https://arxiv.org/abs/1809.06858