Enregistré dans:
Détails bibliographiques
Auteurs principaux: Shen, Jay, Tang, Yifeng, Ferguson, Andrew
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2511.11489
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908653642055680
author Shen, Jay
Tang, Yifeng
Ferguson, Andrew
author_facet Shen, Jay
Tang, Yifeng
Ferguson, Andrew
contents Transformers are the go-to architecture for most data modalities due to their scalability. While they have been applied extensively to molecular property prediction, they do not dominate the field as they do elsewhere. One cause may be the lack of structural biases that effectively capture the relationships between atoms. Here, we investigate attention biases as a simple and natural way to encode structure. Motivated by physical power laws, we propose a family of low-complexity attention biases $b_{ij} = p \log|| \mathbf{r}_i - \mathbf{r}_j||$ which weigh attention probabilities according to interatomic distances. On the QM9 and SPICE datasets, this approach outperforms positional encodings and graph attention while remaining competitive with more complex Gaussian kernel biases. We also show that good attention biases can compensate for a complete ablation of scaled dot-product attention, suggesting a low-cost path toward interpretable molecular transformers.
format Preprint
id arxiv_https___arxiv_org_abs_2511_11489
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Power law attention biases for molecular transformers
Shen, Jay
Tang, Yifeng
Ferguson, Andrew
Computational Physics
Transformers are the go-to architecture for most data modalities due to their scalability. While they have been applied extensively to molecular property prediction, they do not dominate the field as they do elsewhere. One cause may be the lack of structural biases that effectively capture the relationships between atoms. Here, we investigate attention biases as a simple and natural way to encode structure. Motivated by physical power laws, we propose a family of low-complexity attention biases $b_{ij} = p \log|| \mathbf{r}_i - \mathbf{r}_j||$ which weigh attention probabilities according to interatomic distances. On the QM9 and SPICE datasets, this approach outperforms positional encodings and graph attention while remaining competitive with more complex Gaussian kernel biases. We also show that good attention biases can compensate for a complete ablation of scaled dot-product attention, suggesting a low-cost path toward interpretable molecular transformers.
title Power law attention biases for molecular transformers
topic Computational Physics
url https://arxiv.org/abs/2511.11489