Research map / Optimization / Distributed and compressed
Communication-Efficient Distributed Optimization in Networks with Gradient Tracking and Variance Reduction
Boyue Li, Shicong Cen, Yuxin Chen, Yuejie Chi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Network-DANE / Network-SVRG/SARAH instead of DGD and EXTRA | decentralized optimization · synthetic linear regression, kappa=10 and 10^4 | communication rounds | Improved |
| add variance-reduced local updates (Network-SVRG/SARAH) over Network-DANE | Network-DANE · synthetic linear regression | gradient evaluations | Improved |
| increase mixing rounds K per iteration (ablation) | Network-DANE and Network-SVRG · poorly connected graph (alpha_0 = 0.944) | iterations to converge | Improved |
| increase mixing rounds K beyond a threshold (ablation) | Network-DANE and Network-SVRG · poorly connected graph (alpha_0 = 0.944) | communication rounds | No clear effect |
| increase local iterations S for Network-SVRG up to about 0.05m (ablation) | Network-SVRG · synthetic linear regression, kappa=10 | communication rounds | Improved |
| use less connected topology (grid, star, ring vs Erdos-Renyi p=0.3) (ablation) | Network-DANE and Network-SVRG/SARAH · synthetic linear regression, kappa=10 | convergence | Hurt |
| use Network-DANE instead of ADMM and PG-EXTRA | nonsmooth composite optimization · l1-regularized linear regression | convergence | Improved |
| use Network-SVRG/SARAH instead of DSGD and ADMM | neural network training · MNIST, one hidden layer, 20 agents | communication and computation efficiency | Improved |
Closest papers by meaning
- Federated Optimization with Doubly Regularized Drift Correction (ICML 2024 · similarity 0.91)
- Hybrid Decentralized Optimization: Leveraging Both First- and Zeroth-Order Optimizers for Faster Convergence (AAAI 2025 · similarity 0.90)
- Communication-Efficient Topologies for Decentralized Learning with $O(1)$ Consensus Rate (NeurIPS 2022 · similarity 0.90)
- Optimal Algorithms for Smooth and Strongly Convex Distributed Optimization in Networks (ICML 2017 · similarity 0.90)
- Decentralized Non-convex Stochastic Optimization with Heterogeneous Variance (AAAI 2026 · similarity 0.90)
- Towards Faster Decentralized Stochastic Optimization with Communication Compression (ICLR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
