Learning Simpler Language Models with the Differential State Framework

TL;DR

The paper introduces the Differential State Framework, a simple yet effective neural model design that improves long-term memory in language modeling tasks, outperforming complex architectures like LSTM and GRU.

Contribution

It presents the Differential State Framework and the Delta-RNN architecture, demonstrating superior performance and simplicity compared to existing complex recurrent models.

Findings

Delta-RNN outperforms LSTM and GRU in language modeling.

Regularized Delta-RNN performs comparably to state-of-the-art models.

The framework maintains long-term memory with minimal additional parameters.

Abstract

Learning useful information across long time lags is a critical and difficult problem for temporal neural models in tasks such as language modeling. Existing architectures that address the issue are often complex and costly to train. The Differential State Framework (DSF) is a simple and high-performing design that unifies previously introduced gated neural models. DSF models maintain longer-term memory by learning to interpolate between a fast-changing data-driven representation and a slowly changing, implicitly stable state. This requires hardly any more parameters than a classical, simple recurrent network. Within the DSF framework, a new architecture is presented, the Delta-RNN. In language modeling at the word and character levels, the Delta-RNN outperforms popular complex architectures, such as the Long Short Term Memory (LSTM) and the Gated Recurrent Unit (GRU), and, when regularized, performs comparably to several state-of-the-art baselines. At the subword level, the Delta-RNN's performance is comparable to that of complex gated architectures.