Research map / Optimization / Nonconvex and smooth optimization
PMGT-VR: A Decentralized Proximal-Gradient Algorithmic Framework With Variance Reduction
Haishan Ye, Wei Xiong, Tong Zhang
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 stochastic variance-reduced PMGT-SAGA/PMGT-LSVRG instead of full-gradient PG-EXTRA/NIDS | decentralized proximal gradient · a9a L1-logistic regression | suboptimality vs component gradient evaluations: much fewer component gradient evaluations (no numbers given) | Improved |
| use PMGT-VR with multi-consensus instead of PG-EXTRA on a poorly connected network (1-lambda2(W)=0.05) | PMGT-SAGA/PMGT-LSVRG · a9a, 1-lambda2(W)=0.05 | suboptimality convergence: PG-EXTRA converges much slower when 1-lambda2(W) is small | Improved |
| PMGT-VR versus PG-EXTRA/NIDS in the ill-conditioned case sigma=1e-7n | PMGT-SAGA/PMGT-LSVRG · a9a, sigma=1e-7n | suboptimality vs gradient evaluations | Improved |
| PMGT-VR versus PG-EXTRA/NIDS in communication rounds | PMGT-SAGA/PMGT-LSVRG · a9a, sigma=1e-5n | communication rounds | Hurt |
| use PMGT-SAGA instead of PG-EXTRA/NIDS when tau=250 | PMGT-SAGA/PMGT-LSVRG · a9a, all settings | total cost: tau=250 | Improved |
| increase communication cost ratio tau | PMGT-SAGA vs PG-EXTRA · a9a, sigma=1e-5n | relative total cost: similar at tau approx 1300 and tau approx 500 | Hurt |
| increase multi-consensus rounds K per iteration (ablation) | PMGT-LSVRG · a9a, 1-lambda2(W)=0.05, sigma=1e-5n,1e-6n | gradient evaluations to converge: faster at first, no further gain after a threshold while communication rises significantly | Mixed |
| increase K in ill-conditioned case (ablation) | PMGT-LSVRG · a9a, sigma=1e-7n | convergence: impact of multi-consensus is relatively small | No clear effect |
Closest papers by meaning
- Decentralized Sum-of-Nonconvex Optimization (AAAI 2024 · similarity 0.88)
- Jointly Improving the Sample and Communication Complexities in Decentralized Stochastic Minimax Optimization (AAAI 2024 · similarity 0.88)
- Achieving Linear Convergence with Parameter-Free Algorithms in Decentralized Optimization (NeurIPS 2024 · similarity 0.88)
- Decentralized Gradient-Free Methods for Stochastic Non-smooth Non-convex Optimization (AAAI 2024 · similarity 0.87)
- Exploiting Similarity for Computation and Communication-Efficient Decentralized Optimization (ICML 2025 · similarity 0.87)
- Decentralized Convex Finite-Sum Optimization with Better Dependence on Condition Numbers (ICML 2024 · similarity 0.87)
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.
