Randomized Polar Codes for Anytime Distributed Machine Learning

被引:3
作者
Bartan, Burak [1 ]
Pilanci, Mert [1 ]
机构
[1] Stanford Univ, Dept Elect Engn, Stanford, CA 94305 USA
来源
IEEE JOURNAL ON SELECTED AREAS IN INFORMATION THEORY | 2023年 / 4卷
基金
美国国家科学基金会;
关键词
Polar codes; distributed algorithms; randomized sketching; machine learning; large-scale computing; POLARIZATION; COMPUTATION;
D O I
10.1109/JSAIT.2023.3310931
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
We present a novel distributed computing framework that is robust to slow compute nodes, and is capable of both approximate and exact computation of linear operations. The proposed mechanism integrates the concepts of randomized sketching and polar codes in the context of coded computation. We propose a sequential decoding algorithm designed to handle real valued data while maintaining low computational complexity for recovery. Additionally, we provide an anytime estimator that can generate provably accurate estimates even when the set of available node outputs is not decodable. We demonstrate the potential applications of this framework in various contexts, such as large-scale matrix multiplication and black-box optimization. We present the implementation of these methods on a serverless cloud computing system and provide numerical results to demonstrate their scalability in practice, including ImageNet scale computations.
引用
收藏
页码:393 / 404
页数:12
相关论文
共 37 条
[1]   Channel Polarization: A Method for Constructing Capacity-Achieving Codes for Symmetric Binary-Input Memoryless Channels [J].
Arikan, Erdal .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2009, 55 (07) :3051-3073
[2]  
Baharav T, 2018, IEEE INT SYMP INFO, P1993, DOI 10.1109/ISIT.2018.8437549
[3]  
Bartan B, 2019, ANN ALLERTON CONF, P276, DOI [10.1109/allerton.2019.8919767, 10.1109/ALLERTON.2019.8919767]
[4]  
Carreira J., 2018, P WORKSH SYST ML OP, P1
[5]  
Choromanski K, 2018, PR MACH LEARN RES, V80
[6]  
Dean T., 1988, AAAI 88. Seventh National Conference on Artificial Intelligence, P49
[7]  
Deng J, 2009, PROC CVPR IEEE, P248, DOI 10.1109/CVPRW.2009.5206848
[8]  
Didier F, 2009, Arxiv, DOI arXiv:0901.1886
[9]   On the Optimal Recovery Threshold of Coded Matrix Multiplication [J].
Dutta, Sanghamitra ;
Fahim, Mohammad ;
Haddadpour, Farzin ;
Jeong, Haewon ;
Cadambe, Viveck ;
Grover, Pulkit .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2020, 66 (01) :278-301
[10]   Distributed Computations with Layered Resolution [J].
Esfahanizadeh, Homa ;
Cohen, Alejandro ;
Medard, Muriel ;
Shamai , Shlomo .
PROCEEDINGS OF THE 2022 IEEE 11TH INTERNATIONAL CONFERENCE ON CLOUD NETWORKING (IEEE CLOUDNET 2022), 2022, :257-261