Adaptive Fourier Neural Operators: Efficient Token Mixers for Transformers

计算机科学 人工智能 安全性令牌 频域 算法 理论计算机科学 计算机视觉 计算机安全
作者
John Guibas,Morteza Mardani,Zongyi Li,Andrew Tao,Anima Anandkumar,Bryan Catanzaro
出处
期刊:Cornell University - arXiv 被引量:52
标识
DOI:10.48550/arxiv.2111.13587
摘要

Vision transformers have delivered tremendous success in representation learning. This is primarily due to effective token mixing through self attention. However, this scales quadratically with the number of pixels, which becomes infeasible for high-resolution inputs. To cope with this challenge, we propose Adaptive Fourier Neural Operator (AFNO) as an efficient token mixer that learns to mix in the Fourier domain. AFNO is based on a principled foundation of operator learning which allows us to frame token mixing as a continuous global convolution without any dependence on the input resolution. This principle was previously used to design FNO, which solves global convolution efficiently in the Fourier domain and has shown promise in learning challenging PDEs. To handle challenges in visual representation learning such as discontinuities in images and high resolution inputs, we propose principled architectural modifications to FNO which results in memory and computational efficiency. This includes imposing a block-diagonal structure on the channel mixing weights, adaptively sharing weights across tokens, and sparsifying the frequency modes via soft-thresholding and shrinkage. The resulting model is highly parallel with a quasi-linear complexity and has linear memory in the sequence size. AFNO outperforms self-attention mechanisms for few-shot segmentation in terms of both efficiency and accuracy. For Cityscapes segmentation with the Segformer-B3 backbone, AFNO can handle a sequence size of 65k and outperforms other efficient self-attention mechanisms.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
更新
大幅提高文件上传限制,最高150M (2024-4-1)

科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
没有星期八应助Relax采纳,获得10
刚刚
迷你的夏云完成签到 ,获得积分10
1秒前
南柯完成签到,获得积分10
1秒前
2秒前
2秒前
Vincent完成签到,获得积分10
2秒前
李健应助shallow采纳,获得10
3秒前
斯文败类应助科研通管家采纳,获得10
3秒前
传奇3应助科研通管家采纳,获得10
3秒前
上官若男应助科研通管家采纳,获得10
4秒前
大个应助科研通管家采纳,获得10
4秒前
ppg123应助科研通管家采纳,获得30
4秒前
CipherSage应助科研通管家采纳,获得10
4秒前
华仔应助科研通管家采纳,获得10
4秒前
科研通AI2S应助科研通管家采纳,获得20
4秒前
隐形曼青应助科研通管家采纳,获得10
4秒前
邓佳鑫Alan应助科研通管家采纳,获得10
4秒前
Akim应助科研通管家采纳,获得10
4秒前
CodeCraft应助科研通管家采纳,获得10
4秒前
在水一方应助科研通管家采纳,获得10
4秒前
wanci应助科研通管家采纳,获得10
5秒前
隐形曼青应助科研通管家采纳,获得10
5秒前
mikudts发布了新的文献求助10
5秒前
1236应助MoO采纳,获得10
5秒前
5秒前
可爱的函函应助Ray采纳,获得10
8秒前
qin发布了新的文献求助10
9秒前
NexusExplorer应助妞妞采纳,获得10
9秒前
plain发布了新的文献求助10
9秒前
青梅发布了新的文献求助10
10秒前
希望天下0贩的0应助marmota采纳,获得10
13秒前
13秒前
淡淡的白羊完成签到 ,获得积分10
13秒前
13秒前
shallow完成签到,获得积分10
14秒前
15秒前
16秒前
16秒前
专注的芋泥完成签到,获得积分10
17秒前
自私的猫发布了新的文献求助10
18秒前
高分求助中
The late Devonian Standard Conodont Zonation 2000
Nickel superalloy market size, share, growth, trends, and forecast 2023-2030 2000
The Lali Section: An Excellent Reference Section for Upper - Devonian in South China 1500
Very-high-order BVD Schemes Using β-variable THINC Method 870
Mantiden: Faszinierende Lauerjäger Faszinierende Lauerjäger 800
PraxisRatgeber: Mantiden: Faszinierende Lauerjäger 800
A new species of Coccus (Homoptera: Coccoidea) from Malawi 500
热门求助领域 (近24小时)
化学 医学 生物 材料科学 工程类 有机化学 生物化学 物理 内科学 纳米技术 计算机科学 化学工程 复合材料 基因 遗传学 催化作用 物理化学 免疫学 量子力学 细胞生物学
热门帖子
关注 科研通微信公众号,转发送积分 3256267
求助须知:如何正确求助?哪些是违规求助? 2898511
关于积分的说明 8301297
捐赠科研通 2567706
什么是DOI,文献DOI怎么找? 1394629
科研通“疑难数据库(出版商)”最低求助积分说明 652895
邀请新用户注册赠送积分活动 630548