Bayesian Attention Mechanism: A Probabilistic Framework for Positional Encoding and Context Length Extrapolation
Abstract
Transformer-based language models rely on positional encoding (PE) to handle token order and support context length extrapolation. However, existing PE methods lack theoretical clarity and rely on limited evaluation metrics to substantiate their extrapolation claims. We propose the Bayesian Attention Mechanism (BAM), a theoretical framework that formulates positional encoding as a prior within a probabilistic model. BAM unifies existing methods (e.g., NoPE and ALiBi) and motivates a new Generalized Gaussian positional prior that substantially improves long-context generalization. Empirically, BAM enables accurate information retrieval at $500\times$ the training context length, outperforming previous state-of-the-art context length generalization by more than $25\times$ in retrieval accuracy while maintaining comparable perplexity and introducing minimal additional parameters.
Cite
Text
Bianchessi et al. "Bayesian Attention Mechanism: A Probabilistic Framework for Positional Encoding and Context Length Extrapolation." International Conference on Learning Representations, 2026.Markdown
[Bianchessi et al. "Bayesian Attention Mechanism: A Probabilistic Framework for Positional Encoding and Context Length Extrapolation." International Conference on Learning Representations, 2026.](https://mlanthology.org/iclr/2026/bianchessi2026iclr-bayesian/)BibTeX
@inproceedings{bianchessi2026iclr-bayesian,
title = {{Bayesian Attention Mechanism: A Probabilistic Framework for Positional Encoding and Context Length Extrapolation}},
author = {Bianchessi, Arthur S. and Aguirre, Yasmin C. and Barros, Rodrigo C. and Kupssinskü, Lucas S.},
booktitle = {International Conference on Learning Representations},
year = {2026},
url = {https://mlanthology.org/iclr/2026/bianchessi2026iclr-bayesian/}
}