Timezone: »
Poster
Communication-efficient Distributed SGD with Sketching
Nikita Ivkin · Daniel Rothchild · Enayat Ullah · Vladimir Braverman · Ion Stoica · Raman Arora
Tue Dec 10 10:45 AM -- 12:45 PM (PST) @ East Exhibition Hall B + C #81
Large-scale distributed training of neural networks is often limited by network bandwidth, wherein the communication time overwhelms the local computation time. Motivated by the success of sketching methods in sub-linear/streaming algorithms, we introduce Sketched-SGD, an algorithm for carrying out distributed SGD by communicating sketches instead of full gradients. We show that \ssgd has favorable convergence rates on several classes of functions. When considering all communication -- both of gradients and of updated model weights -- Sketched-SGD reduces the amount of communication required compared to other gradient compression methods from $\mathcal{O}(d)$ or $\mathcal{O}(W)$ to $\mathcal{O}(\log d)$, where $d$ is the number of model parameters and $W$ is the number of workers participating in training. We run experiments on a transformer model, an LSTM, and a residual network, demonstrating up to a 40x reduction in total communication cost with no loss in final model performance. We also show experimentally that Sketched-SGD scales to at least 256 workers without increasing communication cost or degrading model performance.
Author Information
Nikita Ivkin (Amazon)
Daniel Rothchild (UC Berkeley)
Enayat Ullah (Johns Hopkins University)
Vladimir Braverman (Johns Hopkins University)
Ion Stoica (UC Berkeley)
Raman Arora (Johns Hopkins University)
More from the Same Authors
-
2021 : TenSet: A Large-scale Program Performance Dataset for Learned Tensor Compilers »
Lianmin Zheng · Ruochen Liu · Junru Shao · Tianqi Chen · Joseph Gonzalez · Ion Stoica · Ameer Haj-Ali -
2021 Spotlight: Coresets for Clustering with Missing Values »
Vladimir Braverman · Shaofeng Jiang · Robert Krauthgamer · Xuan Wu -
2021 : MESA: Offline Meta-RL for Safe Adaptation and Fault Tolerance »
Michael Luo · Ashwin Balakrishna · Brijen Thananjeyan · Suraj Nair · Julian Ibarz · Jie Tan · Chelsea Finn · Ion Stoica · Ken Goldberg -
2022 : Bidirectional Adaptive Communication for Heterogeneous Distributed Learning »
Dmitrii Avdiukhin · Vladimir Braverman · Nikita Ivkin · Sebastian Stich -
2022 : Fifteen-minute Competition Overview Video »
Nathan Drenkow · Raman Arora · Gino Perrotta · Todd Neller · Ryan Gardner · Mykel J Kochenderfer · Jared Markowitz · Corey Lowman · Casey Richardson · Bo Li · Bart Paulhamus · Ashley J Llorens · Andrew Newman -
2022 : From Local to Global: Spectral-Inspired Graph Neural Networks »
Ningyuan Huang · Soledad Villar · Carey E Priebe · Da Zheng · Chengyue Huang · Lin Yang · Vladimir Braverman -
2022 Spotlight: Risk Bounds of Multi-Pass SGD for Least Squares in the Interpolation Regime »
Difan Zou · Jingfeng Wu · Vladimir Braverman · Quanquan Gu · Sham Kakade -
2022 Competition: Reconnaissance Blind Chess: An Unsolved Challenge for Multi-Agent Decision Making Under Uncertainty »
Ryan Gardner · Gino Perrotta · Corey Lowman · Casey Richardson · Andrew Newman · Jared Markowitz · Nathan Drenkow · Bart Paulhamus · Ashley J Llorens · Todd Neller · Raman Arora · Bo Li · Mykel J Kochenderfer -
2022 Poster: Differentially Private Generalized Linear Models Revisited »
Raman Arora · Raef Bassily · Cristóbal Guzmán · Michael Menart · Enayat Ullah -
2022 Poster: The Power and Limitation of Pretraining-Finetuning for Linear Regression under Covariate Shift »
Jingfeng Wu · Difan Zou · Vladimir Braverman · Quanquan Gu · Sham Kakade -
2022 Poster: Adversarial Robustness is at Odds with Lazy Training »
Yunjuan Wang · Enayat Ullah · Poorya Mianjy · Raman Arora -
2022 Poster: Risk Bounds of Multi-Pass SGD for Least Squares in the Interpolation Regime »
Difan Zou · Jingfeng Wu · Vladimir Braverman · Quanquan Gu · Sham Kakade -
2021 Poster: Coresets for Clustering with Missing Values »
Vladimir Braverman · Shaofeng Jiang · Robert Krauthgamer · Xuan Wu -
2021 Poster: The Benefits of Implicit Regularization from SGD in Least Squares Problems »
Difan Zou · Jingfeng Wu · Vladimir Braverman · Quanquan Gu · Dean Foster · Sham Kakade -
2021 Poster: Representing Long-Range Context for Graph Neural Networks with Global Attention »
Zhanghao Wu · Paras Jain · Matthew Wright · Azalia Mirhoseini · Joseph Gonzalez · Ion Stoica -
2021 : Reconnaissance Blind Chess + Q&A »
Ryan Gardner · Gino Perrotta · Corey Lowman · Casey Richardson · Andrew Newman · Jared Markowitz · Nathan Drenkow · Bart Paulhamus · Ashley J Llorens · Todd Neller · Raman Arora · Bo Li · Mykel J Kochenderfer -
2021 Poster: Accommodating Picky Customers: Regret Bound and Exploration Complexity for Multi-Objective Reinforcement Learning »
Jingfeng Wu · Vladimir Braverman · Lin Yang -
2021 Poster: Adversarial Robustness of Streaming Algorithms through Importance Sampling »
Vladimir Braverman · Avinatan Hassidim · Yossi Matias · Mariano Schain · Sandeep Silwal · Samson Zhou -
2020 Poster: Adversarial Robustness of Supervised Sparse Coding »
Jeremias Sulam · Ramchandran Muthukumar · Raman Arora -
2020 Poster: On Convergence and Generalization of Dropout Training »
Poorya Mianjy · Raman Arora -
2019 : Invited Speaker: Ion Stoica »
Ion Stoica -
2019 Poster: Efficient Convex Relaxations for Streaming PCA »
Raman Arora · Teodor Vanislavov Marinov -
2019 Poster: On Differentially Private Graph Sparsification and Applications »
Raman Arora · Jalaj Upadhyay -
2019 Poster: Bandits with Feedback Graphs and Switching Costs »
Raman Arora · Teodor Vanislavov Marinov · Mehryar Mohri -
2018 : Poster Session »
Sujay Sanghavi · Vatsal Shah · Yanyao Shen · Tianchen Zhao · Yuandong Tian · Tomer Galanti · Mufan Li · Gilad Cohen · Daniel Rothchild · Aristide Baratin · Devansh Arpit · Vagelis Papalexakis · Michael Perlmutter · Ashok Vardhan Makkuva · Pim de Haan · Yingyan Lin · Wanmo Kang · Cheolhyoung Lee · Hao Shen · Sho Yaida · Dan Roberts · Nadav Cohen · Philippe Casgrain · Dejiao Zhang · Tengyu Ma · Avinash Ravichandran · Julian Emilio Salazar · Bo Li · Davis Liang · Christopher Wong · Glen Bigan Mbeng · Animesh Garg -
2018 : Poster Session 1 »
Kyle H Ambert · Brandon Araki · Xiya Cao · Sungjoon Choi · Hao(Jackson) Cui · Jonas Degrave · Yaqi Duan · Mattie Fellows · Carlos Florensa · Karan Goel · Aditya Gopalan · Ming-Xu Huang · Jonathan Hunt · Cyril Ibrahim · Brian Ichter · Maximilian Igl · Zheng Tracy Ke · Igor Kiselev · Anuj Mahajan · Arash Mehrjou · Karl Pertsch · Alexandre Piche · Nicholas Rhinehart · Thomas Ringstrom · Reazul Hasan Russel · Oleh Rybkin · Ion Stoica · Sharad Vikram · Angelina Wang · Ting-Han Wei · Abigail H Wen · I-Chen Wu · Zhengwei Wu · Linhai Xie · Dinghan Shen -
2018 Poster: Policy Regret in Repeated Games »
Raman Arora · Michael Dinitz · Teodor Vanislavov Marinov · Mehryar Mohri -
2018 Poster: Streaming Kernel PCA with $\tilde{O}(\sqrt{n})$ Random Features »
Enayat Ullah · Poorya Mianjy · Teodor Vanislavov Marinov · Raman Arora -
2018 Poster: The Physical Systems Behind Optimization Algorithms »
Lin Yang · Raman Arora · Vladimir Braverman · Tuo Zhao -
2018 Poster: Differentially Private Robust Low-Rank Approximation »
Raman Arora · Vladimir Braverman · Jalaj Upadhyay -
2017 : Poster Session »
Tsz Kit Lau · Johannes Maly · Nicolas Loizou · Christian Kroer · Yuan Yao · Youngsuk Park · Reka Agnes Kovacs · Dong Yin · Vlad Zhukov · Woosang Lim · David Barmherzig · Dimitris Metaxas · Bin Shi · Rajan Udwani · William Brendel · Yi Zhou · Vladimir Braverman · Sijia Liu · Eugene Golikov -
2017 Poster: Stochastic Approximation for Canonical Correlation Analysis »
Raman Arora · Teodor Vanislavov Marinov · Poorya Mianjy · Nati Srebro -
2016 Poster: Disease Trajectory Maps »
Peter Schulam · Raman Arora -
2014 Poster: Accelerated Mini-batch Randomized Block Coordinate Descent Method »
Tuo Zhao · Mo Yu · Yiming Wang · Raman Arora · Han Liu -
2013 Poster: Stochastic Optimization of PCA with Capped MSG »
Raman Arora · Andrew Cotter · Nati Srebro -
2009 Poster: On Learning Rotations »
Raman Arora -
2009 Spotlight: On Learning Rotations »
Raman Arora