A Manifold Representation of the Key in Vision Transformers

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Meng, Li, Goodwin, Morten, Yazidi, Anis, Engelstad, Paal
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866929376850870272
author Meng, Li
Goodwin, Morten
Yazidi, Anis
Engelstad, Paal
author_facet Meng, Li
Goodwin, Morten
Yazidi, Anis
Engelstad, Paal
contents Vision Transformers implement multi-head self-attention via stacking multiple attention blocks. The query, key, and value are often intertwined and generated within those blocks via a single, shared linear transformation. This paper explores the concept of disentangling the key from the query and value, and adopting a manifold representation for the key. Our experiments reveal that decoupling and endowing the key with a manifold structure can enhance the model's performance. Specifically, ViT-B exhibits a 0.87% increase in top-1 accuracy, while Swin-T sees a boost of 0.52% in top-1 accuracy on the ImageNet-1K dataset, with eight charts in the manifold key. Our approach also yields positive results in object detection and instance segmentation tasks on the COCO dataset. We establish that these performance gains are not merely due to the simplicity of adding more parameters and computations. Future research may investigate strategies for cutting the budget of such representations and aim for further performance improvements based on our findings.
format Preprint
id arxiv_https___arxiv_org_abs_2402_00534
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle A Manifold Representation of the Key in Vision Transformers
Meng, Li
Goodwin, Morten
Yazidi, Anis
Engelstad, Paal
Computer Vision and Pattern Recognition
Machine Learning
Vision Transformers implement multi-head self-attention via stacking multiple attention blocks. The query, key, and value are often intertwined and generated within those blocks via a single, shared linear transformation. This paper explores the concept of disentangling the key from the query and value, and adopting a manifold representation for the key. Our experiments reveal that decoupling and endowing the key with a manifold structure can enhance the model's performance. Specifically, ViT-B exhibits a 0.87% increase in top-1 accuracy, while Swin-T sees a boost of 0.52% in top-1 accuracy on the ImageNet-1K dataset, with eight charts in the manifold key. Our approach also yields positive results in object detection and instance segmentation tasks on the COCO dataset. We establish that these performance gains are not merely due to the simplicity of adding more parameters and computations. Future research may investigate strategies for cutting the budget of such representations and aim for further performance improvements based on our findings.
title A Manifold Representation of the Key in Vision Transformers
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2402.00534