Linear Log-Normal Attention with Unbiased Concentration

Published: 16 Jan 2024, Last Modified: 05 Mar 2024ICLR 2024 posterEveryoneRevisionsBibTeX
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Neural Networks, Transformers, Self-Attention, Linear Attention, Scalable Transformers, Efficient Attention, Attention with Linear Complexity, Linearized Attention, Self-Attention Analysis
Submission Guidelines: I certify that this submission complies with the submission instructions as described on
TL;DR: The quadratic complexity of the attention limits the scalability of the transformer models. We propose Linear Log-Normal Attention that offers linear complexity while maintaining key features of the original attention mechanism.
Abstract: Transformer models have achieved remarkable results in a wide range of applications. However, their scalability is hampered by the quadratic time and memory complexity of the self-attention mechanism concerning the sequence length. This limitation poses a substantial obstacle when dealing with long documents or high-resolution images. In this work, we study the self-attention mechanism by analyzing the distribution of the attention matrix and its concentration ability. Furthermore, we propose instruments to measure these quantities and introduce a novel self-attention mechanism, Linear Log-Normal Attention, designed to emulate the distribution and concentration behavior of the original self-attention. Our experimental results on popular natural language benchmarks reveal that our proposed Linear Log-Normal Attention outperforms other linearized attention alternatives, offering a promising avenue for enhancing the scalability of transformer models.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
Supplementary Material: zip
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Primary Area: general machine learning (i.e., none of the above)
Submission Number: 5618