Improving Language Modeling using Densely Connected Recurrent Neural Networks

TL;DR

This paper introduces densely connected layers into recurrent neural networks, demonstrating that such architectures can achieve comparable language modeling performance with significantly fewer parameters, especially when connecting only a few layers.

Contribution

The paper presents a novel densely connected RNN architecture that reduces parameters while maintaining performance, highlighting the effectiveness of selective dense connections.

Findings

Achieves similar perplexity with six times fewer parameters

Densely connecting a few layers yields significant perplexity reduction

Outperforms standard stacked LSTM models in efficiency

Abstract

In this paper, we introduce the novel concept of densely connected layers into recurrent neural networks. We evaluate our proposed architecture on the Penn Treebank language modeling task. We show that we can obtain similar perplexity scores with six times fewer parameters compared to a standard stacked 2-layer LSTM model trained with dropout (Zaremba et al. 2014). In contrast with the current usage of skip connections, we show that densely connecting only a few stacked layers with skip connections already yields significant perplexity reductions.