Tensorized Self-Attention: Efficiently Modeling Pairwise and Global Dependencies Together
This addresses efficiency and performance issues in NLP models for researchers and practitioners, though it is incremental as it builds on existing attention mechanisms.
The paper tackles the memory and computation bottlenecks of using tensors for pairwise dependencies in self-attention by proposing Multi-mask Tensorized Self-Attention (MTSA), which achieves state-of-the-art or competitive performance on nine NLP benchmarks with improved efficiency.
Neural networks equipped with self-attention have parallelizable computation, light-weight structure, and the ability to capture both long-range and local dependencies. Further, their expressive power and performance can be boosted by using a vector to measure pairwise dependency, but this requires to expand the alignment matrix to a tensor, which results in memory and computation bottlenecks. In this paper, we propose a novel attention mechanism called "Multi-mask Tensorized Self-Attention" (MTSA), which is as fast and as memory-efficient as a CNN, but significantly outperforms previous CNN-/RNN-/attention-based models. MTSA 1) captures both pairwise (token2token) and global (source2token) dependencies by a novel compatibility function composed of dot-product and additive attentions, 2) uses a tensor to represent the feature-wise alignment scores for better expressive power but only requires parallelizable matrix multiplications, and 3) combines multi-head with multi-dimensional attentions, and applies a distinct positional mask to each head (subspace), so the memory and computation can be distributed to multiple heads, each with sequential information encoded independently. The experiments show that a CNN/RNN-free model based on MTSA achieves state-of-the-art or competitive performance on nine NLP benchmarks with compelling memory- and time-efficiency.