# Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

**Authors:** Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, Yoshua Bengio  
**arXiv:** [1412.3555](https://arxiv.org/abs/1412.3555)  
**Date:** 2014-12-11  
**Summary page:** https://www.summarizepaper.com/en/paper/1412.3555v1/

## Key points

- The paper compares various types of recurrent units in RNNs, focusing on LSTM and GRU units with a gating mechanism.
- Evaluation is based on tasks related to polyphonic music modeling and speech signal modeling.
- Results indicate that LSTM and GRU outperform traditional tanh units.
- GRU performs similarly to LSTM in the evaluation.
- The research was presented at the NIPS 2014 Deep Learning and Representation Learning Workshop, offering insights into the effectiveness of different recurrent units in sequence modeling within neural networks.

## Summary

The paper "Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling" compares various types of recurrent units in RNNs. Specifically, the authors Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio focus on advanced units with a gating mechanism - long short-term memory (LSTM) units and gated recurrent units (GRU). Their evaluation is based on tasks related to polyphonic music modeling and speech signal modeling. The results show that these sophisticated recurrent units outperform traditional tanh units. Additionally, the study finds that GRU performs similarly to LSTM. This research was presented at the NIPS 2014 Deep Learning and Representation Learning Workshop and provides valuable insights into the effectiveness of different recurrent units in sequence modeling within neural networks.