Sampling Foundational Transformer: A Theoretical Perspective

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Nguyen, Viet Anh, Lenhat, Minh, Nguyen, Khoa, Hieu, Duong Duc, Hung, Dao Huu, Hy, Truong Son
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916360717598720
author Nguyen, Viet Anh
Lenhat, Minh
Nguyen, Khoa
Hieu, Duong Duc
Hung, Dao Huu
Hy, Truong Son
author_facet Nguyen, Viet Anh
Lenhat, Minh
Nguyen, Khoa
Hieu, Duong Duc
Hung, Dao Huu
Hy, Truong Son
contents The versatility of self-attention mechanism earned transformers great success in almost all data modalities, with limitations on the quadratic complexity and difficulty of training. To apply transformers across different data modalities, practitioners have to make specific clever data-modality-dependent constructions. In this paper, we propose Sampling Foundational Transformer (SFT) that can work on multiple data modalities (e.g., point cloud, graph, and sequence) and constraints (e.g., rotational-invariant). The existence of such model is important as contemporary foundational modeling requires operability on multiple data sources. For efficiency on large number of tokens, our model relies on our context aware sampling-without-replacement mechanism for both linear asymptotic computational complexity and real inference time gain. For efficiency, we rely on our newly discovered pseudoconvex formulation of transformer layer to increase model's convergence rate. As a model working on multiple data modalities, SFT has achieved competitive results on many benchmarks, while being faster in inference, compared to other very specialized models.
format Preprint
id arxiv_https___arxiv_org_abs_2408_05822
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Sampling Foundational Transformer: A Theoretical Perspective
Nguyen, Viet Anh
Lenhat, Minh
Nguyen, Khoa
Hieu, Duong Duc
Hung, Dao Huu
Hy, Truong Son
Machine Learning
Computer Vision and Pattern Recognition
The versatility of self-attention mechanism earned transformers great success in almost all data modalities, with limitations on the quadratic complexity and difficulty of training. To apply transformers across different data modalities, practitioners have to make specific clever data-modality-dependent constructions. In this paper, we propose Sampling Foundational Transformer (SFT) that can work on multiple data modalities (e.g., point cloud, graph, and sequence) and constraints (e.g., rotational-invariant). The existence of such model is important as contemporary foundational modeling requires operability on multiple data sources. For efficiency on large number of tokens, our model relies on our context aware sampling-without-replacement mechanism for both linear asymptotic computational complexity and real inference time gain. For efficiency, we rely on our newly discovered pseudoconvex formulation of transformer layer to increase model's convergence rate. As a model working on multiple data modalities, SFT has achieved competitive results on many benchmarks, while being faster in inference, compared to other very specialized models.
title Sampling Foundational Transformer: A Theoretical Perspective
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2408.05822