Mher Safaryan

Publication Activity (10 Years)

Years Active: 2020-2024
Publications (10 Years): 20

Top Topics

Variance Reduction

Distributed Learning

Computational Complexity

Steepest Ascent

Top Venues

Publications

Rustem Islamov, Mher Safaryan, Dan Alistarh
AsGrad: A Sharp Unified Analysis of Asynchronous-SGD Algorithms. AISTATS (2024)
Ionut-Vlad Modoranu, Mher Safaryan, Grigory Malinovsky, Eldar Kurtic, Thomas Robert, Peter Richtárik, Dan Alistarh
MicroAdam: Accurate Adaptive Optimization with Low Space Overhead and Provable Convergence. CoRR (2024)
Rustem Islamov, Xun Qian, Slavomír Hanzely, Mher Safaryan, Peter Richtárik
Distributed Newton-Type Methods with Communication Compression and Bernoulli Aggregation. Trans. Mach. Learn. Res. 2023 (2023)
Mher Safaryan, Alexandra Peste, Dan Alistarh
Knowledge Distillation Performs Partial Variance Reduction. NeurIPS (2023)
Mher Safaryan, Alexandra Peste, Dan Alistarh
Knowledge Distillation Performs Partial Variance Reduction. CoRR (2023)
Rustem Islamov, Mher Safaryan, Dan Alistarh
AsGrad: A Sharp Unified Analysis of Asynchronous-SGD Algorithms. CoRR (2023)
Bokun Wang, Mher Safaryan, Peter Richtárik
Theoretically Better and Numerically Faster Distributed Optimization with Smoothness-Aware Quantization Techniques. NeurIPS (2022)
Artavazd Maranjyan, Mher Safaryan, Peter Richtárik
GradSkip: Communication-Accelerated Local Gradient Methods with Better Computational Complexity. CoRR (2022)
Xun Qian, Rustem Islamov, Mher Safaryan, Peter Richtárik
Basis Matters: Better Communication-Efficient Second Order Methods for Federated Learning. AISTATS (2022)
Mher Safaryan, Rustem Islamov, Xun Qian, Peter Richtárik
FedNL: Making Newton-Type Methods Applicable to Federated Learning. ICML (2022)
Rustem Islamov, Xun Qian, Slavomír Hanzely, Mher Safaryan, Peter Richtárik
Distributed Newton-Type Methods with Communication Compression and Bernoulli Aggregation. CoRR (2022)
Xun Qian, Rustem Islamov, Mher Safaryan, Peter Richtárik
Basis Matters: Better Communication-Efficient Second Order Methods for Federated Learning. CoRR (2021)
Mher Safaryan, Rustem Islamov, Xun Qian, Peter Richtárik
FedNL: Making Newton-Type Methods Applicable to Federated Learning. CoRR (2021)
Mher Safaryan, Peter Richtárik
Stochastic Sign Descent Methods: New Algorithms and Better Theory. ICML (2021)
Mher Safaryan, Filip Hanzely, Peter Richtárik
Smoothness Matrices Beat Smoothness Constants: Better Communication Compression Techniques for Distributed Optimization. CoRR (2021)
Mher Safaryan, Filip Hanzely, Peter Richtárik
Smoothness Matrices Beat Smoothness Constants: Better Communication Compression Techniques for Distributed Optimization. NeurIPS (2021)
Bokun Wang, Mher Safaryan, Peter Richtárik
Smoothness-Aware Quantization Techniques. CoRR (2021)
Alyazeed Albasyoni, Mher Safaryan, Laurent Condat, Peter Richtárik
Optimal Gradient Compression for Distributed and Federated Learning. CoRR (2020)
Mher Safaryan, Egor Shulgin, Peter Richtárik
Uncertainty Principle for Communication Compression in Distributed and Federated Learning and the Search for an Optimal Compressor. CoRR (2020)
Aleksandr Beznosikov, Samuel Horváth, Peter Richtárik, Mher Safaryan
On Biased Compression for Distributed Learning. CoRR (2020)