Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox
Picture for Suvrit Sra

Suvrit Sra

Can Single-Shuffle SGD be Better than Reshuffling SGD and GD?


Mar 12, 2021
Chulhee Yun, Suvrit Sra, Ali Jadbabaie

Figure 1 for Can Single-Shuffle SGD be Better than Reshuffling SGD and GD?
Figure 2 for Can Single-Shuffle SGD be Better than Reshuffling SGD and GD?
* 26 pages, 2 figures 
  Access Paper or Ask Questions

Provably Efficient Algorithms for Multi-Objective Competitive RL


Feb 05, 2021
Tiancheng Yu, Yi Tian, Jingzhao Zhang, Suvrit Sra

Figure 1 for Provably Efficient Algorithms for Multi-Objective Competitive RL
Figure 2 for Provably Efficient Algorithms for Multi-Objective Competitive RL
Figure 3 for Provably Efficient Algorithms for Multi-Objective Competitive RL
Figure 4 for Provably Efficient Algorithms for Multi-Objective Competitive RL

  Access Paper or Ask Questions

Why do classifier accuracies show linear trends under distribution shift?


Dec 31, 2020
Horia Mania, Suvrit Sra

Figure 1 for Why do classifier accuracies show linear trends under distribution shift?
Figure 2 for Why do classifier accuracies show linear trends under distribution shift?
Figure 3 for Why do classifier accuracies show linear trends under distribution shift?
Figure 4 for Why do classifier accuracies show linear trends under distribution shift?
* 14 pages, 11 figures 
  Access Paper or Ask Questions

Provably Efficient Online Agnostic Learning in Markov Games


Oct 28, 2020
Yi Tian, Yuanhao Wang, Tiancheng Yu, Suvrit Sra

Figure 1 for Provably Efficient Online Agnostic Learning in Markov Games
Figure 2 for Provably Efficient Online Agnostic Learning in Markov Games

  Access Paper or Ask Questions

Coping with Label Shift via Distributionally Robust Optimisation


Oct 23, 2020
Jingzhao Zhang, Aditya Menon, Andreas Veit, Srinadh Bhojanapalli, Sanjiv Kumar, Suvrit Sra

Figure 1 for Coping with Label Shift via Distributionally Robust Optimisation
Figure 2 for Coping with Label Shift via Distributionally Robust Optimisation
Figure 3 for Coping with Label Shift via Distributionally Robust Optimisation
Figure 4 for Coping with Label Shift via Distributionally Robust Optimisation

  Access Paper or Ask Questions

Contrastive Learning with Hard Negative Samples


Oct 09, 2020
Joshua Robinson, Ching-Yao Chuang, Suvrit Sra, Stefanie Jegelka

Figure 1 for Contrastive Learning with Hard Negative Samples
Figure 2 for Contrastive Learning with Hard Negative Samples
Figure 3 for Contrastive Learning with Hard Negative Samples
Figure 4 for Contrastive Learning with Hard Negative Samples

  Access Paper or Ask Questions

Towards Minimax Optimal Reinforcement Learning in Factored Markov Decision Processes


Jun 24, 2020
Yi Tian, Jian Qian, Suvrit Sra

* 54 pages 
  Access Paper or Ask Questions

SGD with shuffling: optimal rates without component convexity and large epoch requirements


Jun 22, 2020
Kwangjun Ahn, Chulhee Yun, Suvrit Sra

Figure 1 for SGD with shuffling: optimal rates without component convexity and large epoch requirements
Figure 2 for SGD with shuffling: optimal rates without component convexity and large epoch requirements
* 53 pages; supersedes the preprint arXiv:2004.08657; v2 corrects an erroneous claim about SingleShuffle and newly adds Theorem 24 and Appendix F for SingleShuffle 
  Access Paper or Ask Questions

Stochastic Optimization with Non-stationary Noise


Jun 09, 2020
Jingzhao Zhang, Hongzhou Lin, Subhro Das, Suvrit Sra, Ali Jadbabaie

Figure 1 for Stochastic Optimization with Non-stationary Noise
Figure 2 for Stochastic Optimization with Non-stationary Noise
Figure 3 for Stochastic Optimization with Non-stationary Noise
Figure 4 for Stochastic Optimization with Non-stationary Noise

  Access Paper or Ask Questions

On Tight Convergence Rates of Without-replacement SGD


Apr 18, 2020
Kwangjun Ahn, Suvrit Sra

Figure 1 for On Tight Convergence Rates of Without-replacement SGD
* 12 pages 
  Access Paper or Ask Questions

Strength from Weakness: Fast Learning Using Weak Supervision


Feb 19, 2020
Joshua Robinson, Stefanie Jegelka, Suvrit Sra

Figure 1 for Strength from Weakness: Fast Learning Using Weak Supervision
Figure 2 for Strength from Weakness: Fast Learning Using Weak Supervision
Figure 3 for Strength from Weakness: Fast Learning Using Weak Supervision
Figure 4 for Strength from Weakness: Fast Learning Using Weak Supervision
* 21 pages, 8 figures 
  Access Paper or Ask Questions

On Complexity of Finding Stationary Points of Nonsmooth Nonconvex Functions


Feb 16, 2020
Jingzhao Zhang, Hongzhou Lin, Stefanie Jegelka, Ali Jadbabaie, Suvrit Sra

Figure 1 for On Complexity of Finding Stationary Points of Nonsmooth Nonconvex Functions
Figure 2 for On Complexity of Finding Stationary Points of Nonsmooth Nonconvex Functions
Figure 3 for On Complexity of Finding Stationary Points of Nonsmooth Nonconvex Functions

  Access Paper or Ask Questions

From Nesterov's Estimate Sequence to Riemannian Acceleration


Jan 24, 2020
Kwangjun Ahn, Suvrit Sra

Figure 1 for From Nesterov's Estimate Sequence to Riemannian Acceleration
* 30 pages 
  Access Paper or Ask Questions

Learning Adversarial MDPs with Bandit Feedback and Unknown Transition


Jan 07, 2020
Chi Jin, Tiancheng Jin, Haipeng Luo, Suvrit Sra, Tiancheng Yu

* Improved the algorithm with a tighter confidence set 
  Access Paper or Ask Questions

Why ADAM Beats SGD for Attention Models


Dec 06, 2019
Jingzhao Zhang, Sai Praneeth Karimireddy, Andreas Veit, Seungyeon Kim, Sashank J Reddi, Sanjiv Kumar, Suvrit Sra

Figure 1 for Why ADAM Beats SGD for Attention Models
Figure 2 for Why ADAM Beats SGD for Attention Models
Figure 3 for Why ADAM Beats SGD for Attention Models
Figure 4 for Why ADAM Beats SGD for Attention Models

  Access Paper or Ask Questions

Nonconvex stochastic optimization on manifolds via Riemannian Frank-Wolfe methods


Oct 09, 2019
Melanie Weber, Suvrit Sra

Figure 1 for Nonconvex stochastic optimization on manifolds via Riemannian Frank-Wolfe methods
Figure 2 for Nonconvex stochastic optimization on manifolds via Riemannian Frank-Wolfe methods
Figure 3 for Nonconvex stochastic optimization on manifolds via Riemannian Frank-Wolfe methods
Figure 4 for Nonconvex stochastic optimization on manifolds via Riemannian Frank-Wolfe methods

  Access Paper or Ask Questions

Efficient Policy Learning for Non-Stationary MDPs under Adversarial Manipulation


Aug 21, 2019
Tiancheng Yu, Suvrit Sra

* There is a problem in the Theorem 1. We will try to fix it and update a new version 
  Access Paper or Ask Questions

Near Optimal Stratified Sampling


Jul 26, 2019
Tiancheng Yu, Xiyu Zhai, Suvrit Sra

Figure 1 for Near Optimal Stratified Sampling
Figure 2 for Near Optimal Stratified Sampling
* We have discovered a mistake in the main result. The quantity on the RHS of (3) is not equal to the variance of estimator (2) when the sampling rule is designed adaptively as we do. There will be further cross-product terms which are now dominant terms. Therefore, although our bound is correct for (3), it no longer implies bound of the variance of (2) 
  Access Paper or Ask Questions

Are deep ResNets provably better than linear predictors?


Jul 09, 2019
Chulhee Yun, Suvrit Sra, Ali Jadbabaie

Figure 1 for Are deep ResNets provably better than linear predictors?
* 17 pages 
  Access Paper or Ask Questions

Flexible Modeling of Diversity with Strongly Log-Concave Distributions


Jun 12, 2019
Joshua Robinson, Suvrit Sra, Stefanie Jegelka

Figure 1 for Flexible Modeling of Diversity with Strongly Log-Concave Distributions
Figure 2 for Flexible Modeling of Diversity with Strongly Log-Concave Distributions
Figure 3 for Flexible Modeling of Diversity with Strongly Log-Concave Distributions
Figure 4 for Flexible Modeling of Diversity with Strongly Log-Concave Distributions

  Access Paper or Ask Questions

Analysis of Gradient Clipping and Adaptive Scaling with a Relaxed Smoothness Condition


May 28, 2019
Jingzhao Zhang, Tianxing He, Suvrit Sra, Ali Jadbabaie

Figure 1 for Analysis of Gradient Clipping and Adaptive Scaling with a Relaxed Smoothness Condition
Figure 2 for Analysis of Gradient Clipping and Adaptive Scaling with a Relaxed Smoothness Condition
Figure 3 for Analysis of Gradient Clipping and Adaptive Scaling with a Relaxed Smoothness Condition
Figure 4 for Analysis of Gradient Clipping and Adaptive Scaling with a Relaxed Smoothness Condition

  Access Paper or Ask Questions

Escaping Saddle Points with Adaptive Gradient Methods


Jan 26, 2019
Matthew Staib, Sashank J. Reddi, Satyen Kale, Sanjiv Kumar, Suvrit Sra

Figure 1 for Escaping Saddle Points with Adaptive Gradient Methods
Figure 2 for Escaping Saddle Points with Adaptive Gradient Methods

  Access Paper or Ask Questions

Deep-RBF Networks Revisited: Robust Classification with Rejection


Dec 07, 2018
Pourya Habib Zadeh, Reshad Hosseini, Suvrit Sra

Figure 1 for Deep-RBF Networks Revisited: Robust Classification with Rejection
Figure 2 for Deep-RBF Networks Revisited: Robust Classification with Rejection
Figure 3 for Deep-RBF Networks Revisited: Robust Classification with Rejection
Figure 4 for Deep-RBF Networks Revisited: Robust Classification with Rejection

  Access Paper or Ask Questions

R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with Curvature Independent Rate


Nov 28, 2018
Jingzhao Zhang, Hongyi Zhang, Suvrit Sra

Figure 1 for R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with Curvature Independent Rate
Figure 2 for R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with Curvature Independent Rate
* arXiv admin note: text overlap with arXiv:1605.07147 
  Access Paper or Ask Questions

Learning Determinantal Point Processes by Sampling Inferred Negatives


Nov 02, 2018
Zelda Mariet, Mike Gartrell, Suvrit Sra

Figure 1 for Learning Determinantal Point Processes by Sampling Inferred Negatives
Figure 2 for Learning Determinantal Point Processes by Sampling Inferred Negatives
Figure 3 for Learning Determinantal Point Processes by Sampling Inferred Negatives
Figure 4 for Learning Determinantal Point Processes by Sampling Inferred Negatives

  Access Paper or Ask Questions