Timezone: »
Decentralization is a promising method of scaling up parallel machine learning systems. In this paper, we provide a tight lower bound on the iteration complexity for such methods in a stochastic non-convex setting. Our lower bound reveals a theoretical gap in known convergence rates of many existing decentralized training algorithms, such as D-PSGD. We prove by construction this lower bound is tight and achievable. Motivated by our insights, we further propose DeTAG, a practical gossip-style decentralized algorithm that achieves the lower bound with only a logarithm gap. Empirically, we compare DeTAG with other decentralized algorithms on image classification tasks, and we show DeTAG enjoys faster convergence compared to baselines, especially on unshuffled data and in sparse networks.
Author Information
Yucheng Lu (Cornell University)
Christopher De Sa (Cornell)
Related Events (a corresponding poster, oral, or spotlight)
-
2021 Oral: Optimal Complexity in Decentralized Training »
Tue. Jul 20th 12:00 -- 12:20 PM Room
More from the Same Authors
-
2023 : CD-GraB: Coordinating Distributed Example Orders for Provably Accelerated Training »
A. Feder Cooper · Wentao Guo · Duc Khiem Pham · Tiancheng Yuan · Charlie Ruan · Yucheng Lu · Chris De Sa -
2023 Poster: CocktailSGD: Fine-tuning Foundation Models over 500Mbps Networks »
Jue Wang · Yucheng Lu · Binhang Yuan · Beidi Chen · Percy Liang · Chris De Sa · Christopher Re · Ce Zhang -
2023 Poster: STEP: Learning N:M Structured Sparsity Masks from Scratch with Precondition »
Yucheng Lu · Shivani Agrawal · Suvinay Subramanian · Oleg Rybakov · Chris De Sa · Amir Yazdanbakhsh -
2022 : MCTensor: A High-Precision Deep Learning Library with Multi-Component Floating-Point »
Tao Yu · Wentao Guo · Canal Li · Tiancheng Yuan · Christopher De Sa -
2022 : Riemannian Residual Neural Networks »
Isay Katsman · Eric Chen · Sidhanth Holalkere · Aaron Lou · Ser Nam Lim · Christopher De Sa -
2022 Poster: Low-Precision Stochastic Gradient Langevin Dynamics »
Ruqi Zhang · Andrew Wilson · Christopher De Sa -
2022 Spotlight: Low-Precision Stochastic Gradient Langevin Dynamics »
Ruqi Zhang · Andrew Wilson · Christopher De Sa -
2021 Poster: Variance Reduced Training with Stratified Sampling for Forecasting Models »
Yucheng Lu · Youngsuk Park · Lifan Chen · Yuyang Wang · Christopher De Sa · Dean Foster -
2021 Spotlight: Variance Reduced Training with Stratified Sampling for Forecasting Models »
Yucheng Lu · Youngsuk Park · Lifan Chen · Yuyang Wang · Christopher De Sa · Dean Foster -
2021 Poster: Low-Precision Reinforcement Learning: Running Soft Actor-Critic in Half Precision »
Johan Björck · Xiangyu Chen · Christopher De Sa · Carla Gomes · Kilian Weinberger -
2021 Spotlight: Low-Precision Reinforcement Learning: Running Soft Actor-Critic in Half Precision »
Johan Björck · Xiangyu Chen · Christopher De Sa · Carla Gomes · Kilian Weinberger -
2020 Poster: Moniqua: Modulo Quantized Communication in Decentralized SGD »
Yucheng Lu · Christopher De Sa -
2020 Poster: Differentiating through the Fréchet Mean »
Aaron Lou · Isay Katsman · Qingxuan Jiang · Serge Belongie · Ser Nam Lim · Christopher De Sa -
2019 Poster: Distributed Learning with Sublinear Communication »
Jayadev Acharya · Christopher De Sa · Dylan Foster · Karthik Sridharan -
2019 Oral: Distributed Learning with Sublinear Communication »
Jayadev Acharya · Christopher De Sa · Dylan Foster · Karthik Sridharan -
2019 Poster: SWALP : Stochastic Weight Averaging in Low Precision Training »
Guandao Yang · Tianyi Zhang · Polina Kirichenko · Junwen Bai · Andrew Wilson · Christopher De Sa -
2019 Poster: A Kernel Theory of Modern Data Augmentation »
Tri Dao · Albert Gu · Alexander J Ratner · Virginia Smith · Christopher De Sa · Christopher Re -
2019 Poster: Improving Neural Network Quantization without Retraining using Outlier Channel Splitting »
Ritchie Zhao · Yuwei Hu · Jordan Dotzel · Christopher De Sa · Zhiru Zhang -
2019 Oral: SWALP : Stochastic Weight Averaging in Low Precision Training »
Guandao Yang · Tianyi Zhang · Polina Kirichenko · Junwen Bai · Andrew Wilson · Christopher De Sa -
2019 Oral: Improving Neural Network Quantization without Retraining using Outlier Channel Splitting »
Ritchie Zhao · Yuwei Hu · Jordan Dotzel · Christopher De Sa · Zhiru Zhang -
2019 Oral: A Kernel Theory of Modern Data Augmentation »
Tri Dao · Albert Gu · Alexander J Ratner · Virginia Smith · Christopher De Sa · Christopher Re -
2018 Poster: Minibatch Gibbs Sampling on Large Graphical Models »
Christopher De Sa · Vincent Chen · Wong -
2018 Oral: Minibatch Gibbs Sampling on Large Graphical Models »
Christopher De Sa · Vincent Chen · Wong -
2018 Poster: Representation Tradeoffs for Hyperbolic Embeddings »
Frederic Sala · Christopher De Sa · Albert Gu · Christopher Re -
2018 Oral: Representation Tradeoffs for Hyperbolic Embeddings »
Frederic Sala · Christopher De Sa · Albert Gu · Christopher Re