On the Effectiveness of Low-Rank Matrix Factorization for LSTM Model Compression
This work addresses computational bottlenecks for NLP practitioners using LSTMs, but it is incremental as it applies an existing compression method to specific model components.
The paper tackles the computational inefficiency of LSTM networks in NLP tasks by applying low-rank matrix factorization to compress recurrences, finding that additive recurrence is more important than multiplicative recurrence and demonstrating effectiveness in language models and ELMo biLSTM layers across downstream tasks.
Despite their ubiquity in NLP tasks, Long Short-Term Memory (LSTM) networks suffer from computational inefficiencies caused by inherent unparallelizable recurrences, which further aggravates as LSTMs require more parameters for larger memory capacity. In this paper, we propose to apply low-rank matrix factorization (MF) algorithms to different recurrences in LSTMs, and explore the effectiveness on different NLP tasks and model components. We discover that additive recurrence is more important than multiplicative recurrence, and explain this by identifying meaningful correlations between matrix norms and compression performance. We compare our approach across two settings: 1) compressing core LSTM recurrences in language models, 2) compressing biLSTM layers of ELMo evaluated in three downstream NLP tasks.