Compression of Recurrent Neural Networks for Efficient Language Modeling

Abstract

Recurrent neural networks have proved to be an effective method for statistical language modeling. However, in practice their memory and run-time complexity are usually too large to be implemented in real-time offline mobile applications. In this paper we consider several compression techniques for recurrent neural networks including Long-Short Term Memory models. We make particular attention to the high-dimensional output problem caused by the very large vocabulary size. We focus on effective compression methods in the context of their exploitation on devices: pruning, quantization, and matrix decomposition approaches (low-rank factorization and tensor train decomposition, in particular). For each model we investigate the trade-off between its size, suitability for fast inference and perplexity. We propose a general pipeline for applying the most suitable methods to compress recurrent neural networks for language modeling. It has been shown in the experimental study with the Penn Treebank (PTB) dataset that the most efficient results in terms of speed and compression-perplexity balance are obtained by matrix decomposition techniques.

arXiv comment25 pages, 3 tables, 4 figures

References in corpus (15)

• Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
• Recurrent Neural Network Regularization
• Character-Aware Neural Language Models
• A Theoretically Grounded Application of Dropout in Recurrent Neural Networks
• Variational Dropout and the Local Reparameterization Trick
• Variational Dropout Sparsifies Deep Neural Networks
• Unitary Evolution Recurrent Neural Networks
• Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling
• Structured Bayesian Pruning via Log-Normal Multiplicative Noise
• Compressing Recurrent Neural Network with Tensor Train
• Ultimate tensorization: compressing convolutional and FC layers alike
• Trellis Networks for Sequence Modeling
• On the Compression of Recurrent Neural Networks with an Application to LVCSR acoustic modeling for Embedded Speech Recognition
• Neural Networks Compression for Language Modeling
• Bayesian Sparsification of Recurrent Neural Networks

Cited by in corpus (8)

• Long short-term memory networks for proton dose calculation in highly heterogeneous tissues
• Preferences Prediction using a Gallery of Mobile Device based on Scene Recognition and Object Detection
• Restricted Recurrent Neural Networks
• Anchor & Transform: Learning Sparse Embeddings for Large Vocabularies
• Compressing Language Models using Doped Kronecker Products
• Spectral Pruning for Recurrent Neural Networks
• Doping: A technique for efficient compression of LSTM models using sparse structured additive matrices
• Channel Pruning Guided by Spatial and Channel Attention for DNNs in Intelligent Edge Computing