Juncheng Li

Bosheng Qin, Juncheng Li, Siliang Tang et al.

Many studies have been conducted to improve the efficiency of Transformer from quadric to linear. Among them, the low-rank-based methods aim to learn the projection matrices to compress the sequence length. However, the projection matrices are fixed once they have been learned, which compress sequence length with dedicated coefficients for tokens in the same position. Adopting such input-invariant projections ignores the fact that the most informative part of a sequence varies from sequence to sequence, thus failing to preserve the most useful information that lies in varied positions. In addition, previous efficient Transformers only focus on the influence of sequence length while neglecting the effect of hidden state dimension. To address the aforementioned problems, we present an efficient yet effective attention mechanism, namely the Dynamic Bilinear Low-Rank Attention (DBA), which compresses the sequence length by input-sensitive dynamic projection matrices and achieves linear time and space complexity by jointly optimizing the sequence length and hidden state dimension while maintaining state-of-the-art performance. Specifically, we first theoretically demonstrate that the sequence length can be compressed non-destructively from a novel perspective of information theory, with compression matrices dynamically determined by the input sequence. Furthermore, we show that the hidden state dimension can be approximated by extending the Johnson-Lindenstrauss lemma, optimizing the attention in bilinear form. Theoretical analysis shows that DBA is proficient in capturing high-order relations in cross-attention problems. Experiments over tasks with diverse sequence length conditions show that DBA achieves state-of-the-art performance compared with various strong baselines while maintaining less memory consumption with higher speed.

Zhisheng Lu, Juncheng Li, Hong Liu et al.

Single image super-resolution (SISR) has witnessed great strides with the development of deep learning. However, most existing studies focus on building more complex networks with a massive number of layers. Recently, more and more researchers start to explore the application of Transformer in computer vision tasks. However, the heavy computational cost and high GPU memory occupation of the vision Transformer cannot be ignored. In this paper, we propose a novel Efficient Super-Resolution Transformer (ESRT) for SISR. ESRT is a hybrid model, which consists of a Lightweight CNN Backbone (LCB) and a Lightweight Transformer Backbone (LTB). Among them, LCB can dynamically adjust the size of the feature map to extract deep features with a low computational cost. LTB is composed of a series of Efficient Transformers (ET), which occupies a small GPU memory occupation, thanks to the specially designed Efficient Multi-Head Attention (EMHA). Extensive experiments show that ESRT achieves competitive results with low computational costs. Compared with the original Transformer which occupies 16,057M GPU memory, ESRT only occupies 4,191M GPU memory. All codes are available at https://github.com/luissen/ESRT.