1 / 8

LSTM Neural Networks for Language Modeling

LSTM Neural Networks for Language Modeling. Martin Sundermeyer , Ralf Schl¨uter , and Hermann Ney Human Language Technology and Pattern Recognition, Computer Science Department, RWTH Aachen University, Aachen, Germany. INTERSPEECH 2012. 報告者：郝柏翰. Outline. Introduction

adeola
Download Presentation

LSTM Neural Networks for Language Modeling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. LSTM Neural Networks for Language Modeling Martin Sundermeyer, Ralf Schl¨uter, and Hermann NeyHuman Language Technology and Pattern Recognition, Computer Science Department, RWTH Aachen University, Aachen, Germany INTERSPEECH 2012 報告者：郝柏翰

  2. Outline • Introduction • Long Short Term Memory Neural Networks • Experiments • Conclusion

  3. Introduction • Neural networks have become increasingly popular for the task of language modeling. Because neural network LMs overcome a major drawback of backing-off models. • Unfortunately, recurrent neural networks are hard to train using back-propagation through time(BPTT). The main difficulty lies in the well-known vanishing gradient problem which means that the gradient that is propagated back through the network either decays or grows exponentially.

  4. LSTM • The Nodes(blue) are used to control the information flow from the network into the cell and from the cell back into the network. Without modified BPTT algorithm • To avoid this scaling effect, the authors re-designed the unit of a neural network in such a way that its corresponding scaling factor is fixed to one.

  5. References

  6. LSTM • To reduce the computational effort it was proposed to split the words into a set of disjoint word classes. Then the probability can be factorized as follows: • Where , and is the class of word .

  7. Experiments • Corpus • We found that the perplexity of the models was constantly lower by about 8% compared to the standard recurrent neural network. The perplexities we obtained with the sigmoidal recurrent network closely match those obtained with the rnnlmtoolkit.

  8. Experiments

More Related