Distributed Optimization

Improved Convergence in Parameter-Agnostic Error Feedback through Momentum

We study normalized error feedback algorithms with momentum and parameter-agnostic stepsizes, eliminating the need for problem-dependent tuning while achieving competitive convergence rates.

Nov 1, 2025

Bernoulli-LoRA: A Theoretical Framework for Randomized Low-Rank Adaptation

We introduce Bernoulli-LoRA, a theoretical framework for randomized Low-Rank Adaptation that unifies existing approaches and provides convergence guarantees for various optimization methods.

Aug 1, 2025

MARINA-P: Superior Performance in Non-smooth Federated Optimization with Adaptive Stepsizes

We extend MARINA-P algorithm to non-smooth federated optimization, providing the first theoretical analysis with server-to-worker compression and adaptive stepsizes while achieving optimal convergence rates.

Dec 22, 2024

Cohort Squeeze: Beyond a Single Communication Round per Cohort in Cross-Device Federated Learning

We propose a novel federated learning approach that allows multiple communication rounds per cohort, achieving up to 74% reduction in total communication costs through a new stochastic proximal point method variant.

Jun 1, 2024

Don't Compress Gradients in Random Reshuffling: Compress Gradient Differences

This work introduces novel methods combining random reshuffling with gradient compression for distributed and federated learning, providing theoretical analysis and practical improvements over existing approaches.

Jun 14, 2022

3PC: Three Point Compressors for Communication-Efficient Distributed Training and a Better Theory for Lazy Aggregation

We introduce three point compressors (3PC), a novel framework unifying and improving communication-efficient distributed optimization methods. Presented as a Poster at ICML 2022.

Feb 2, 2022

EF21 with Bells & Whistles: Practical Algorithmic Extensions of Modern Error Feedback

Six practical algorithmic extensions of the EF21 error feedback method for communication-efficient distributed learning, with strong convergence guarantees.

Oct 7, 2021

EF21: A New, Simpler, Theoretically Better, and Practically Faster Error Feedback

We propose EF21, a novel approach to error feedback offering a better theoretical rate and strong empirical results. Presented as an Oral + Poster at NeurIPS 2021.

Jun 9, 2021