Bi-Directional Block Self-Attention for Fast and Memory-Efficient Sequence Modeling

TL;DR

Bi-BloSAN is a novel sequence encoding model that combines intra- and inter-block self-attention to achieve RNN/CNN-like memory efficiency while maintaining SAN's advantages, excelling across NLP tasks.

Contribution

Introduces Bi-BloSAN, a bi-directional block self-attention network that reduces memory usage and improves efficiency in sequence modeling compared to existing SAN, RNN, and CNN models.

Findings

Achieves or surpasses state-of-the-art accuracy on nine NLP benchmarks.

Demonstrates better efficiency-memory trade-off than existing RNN, CNN, and SAN models.

Effectively models both local and long-range dependencies with reduced memory requirements.

Abstract

Recurrent neural networks (RNN), convolutional neural networks (CNN) and self-attention networks (SAN) are commonly used to produce context-aware representations. RNN can capture long-range dependency but is hard to parallelize and not time-efficient. CNN focuses on local dependency but does not perform well on some tasks. SAN can model both such dependencies via highly parallelizable computation, but memory requirement grows rapidly in line with sequence length. In this paper, we propose a model, called "bi-directional block self-attention network (Bi-BloSAN)", for RNN/CNN-free sequence encoding. It requires as little memory as RNN but with all the merits of SAN. Bi-BloSAN splits the entire sequence into blocks, and applies an intra-block SAN to each block for modeling local context, then applies an inter-block SAN to the outputs for all blocks to capture long-range dependency. Thus, each SAN only needs to process a short sequence, and only a small amount of memory is required. Additionally, we use feature-level attention to handle the variation of contexts around the same word, and use forward/backward masks to encode temporal order information. On nine benchmark datasets for different NLP tasks, Bi-BloSAN achieves or improves upon state-of-the-art accuracy, and shows better efficiency-memory trade-off than existing RNN/CNN/SAN.