Item request has been placed!
×
Item request cannot be made.
×
Processing Request
Learning a Fourier Transform for Linear Relative Positional Encodings in Transformers
Item request has been placed!
×
Item request cannot be made.
×
Processing Request
- المؤلفون: Choromanski, Krzysztof Marcin; Li, Shanda; Likhosherstov, Valerii; Dubey, Kumar Avinava; Luo, Shengjie; He, Di; Yang, Yiming; Sarlos, Tamas; Weingarten, Thomas; Weller, Adrian
- نوع التسجيلة:
Electronic Resource
- الدخول الالكتروني :
http://arxiv.org/abs/2302.01925
- معلومة اضافية
- Publisher Information:
2023-02-03 2024-04-03
- نبذة مختصرة :
We propose a new class of linear Transformers called FourierLearner-Transformers (FLTs), which incorporate a wide range of relative positional encoding mechanisms (RPEs). These include regular RPE techniques applied for sequential data, as well as novel RPEs operating on geometric data embedded in higher-dimensional Euclidean spaces. FLTs construct the optimal RPE mechanism implicitly by learning its spectral representation. As opposed to other architectures combining efficient low-rank linear attention with RPEs, FLTs remain practical in terms of their memory usage and do not require additional assumptions about the structure of the RPE mask. Besides, FLTs allow for applying certain structural inductive bias techniques to specify masking strategies, e.g. they provide a way to learn the so-called local RPEs introduced in this paper and give accuracy gains as compared with several other linear Transformers for language modeling. We also thoroughly test FLTs on other data modalities and tasks, such as image classification, 3D molecular modeling, and learnable optimizers. To the best of our knowledge, for 3D molecular data, FLTs are the first Transformer architectures providing linear attention and incorporating RPE masking.
Comment: AISTATS 2024
- الموضوع:
- Other Numbers:
COO oai:arXiv.org:2302.01925
1381600018
- Contributing Source:
CORNELL UNIV
From OAIster®, provided by the OCLC Cooperative.
- الرقم المعرف:
edsoai.on1381600018
HoldingsOnline
No Comments.