The paper “Relative positional encoding for transformers with linear complexity” Liutkus et al. has been accepted for presentation at ICML 2021 as long paper (3% of total number of submissions). In 2021, of the 5513 articles submitted, only 1184 were accepted in short presentation (21.5%) and 166 in long presentation (3%).
Titre: Relative positional encoding for transformers with linear complexity
Auteurs: A. Liutkus and O. Cifka and S. Wu and U. Simsekli and Y. Yang and G. Richard,
Abstract: Recent advances in Transformer models allow for unprecedented sequence lengths, due to linear space and time complexity. In the meantime, relative positional encoding (RPE) was proposed as beneficial for classical Transformers and consists in exploiting lags instead of absolute positions for inference. Still, RPE is not available for the recent linear-variants of the Transformer, because it requires the explicit computation of the attention matrix, which is precisely what is avoided by such methods. In this paper, we bridge this gap and present \textit{Stochastic Positional Encoding} as a way to generate PE that can be used as a replacement to the classical additive (sinusoidal) PE and provably behaves like RPE. The main theoretical contribution is to make a connection between positional encoding and cross-covariance structures of correlated Gaussian processes. We illustrate the performance of our approach on the Long-Range Arena benchmark and on music generation.
