Research map / Optimization / Nonconvex and smooth optimization
Finding Local Minima Efficiently in Decentralized Optimization
Wenhan Xian, Heng Huang
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 PEDESTAL instead of first-order decentralized baselines (D-PSGD, GTDSGD, D-GET, D-SPIDER-SFO, GTHSGD) | decentralized matrix sensing · Matrix sensing d=50/100, ring/toroidal/exponential, random data | loss reaching global minimum | Improved |
| use PEDESTAL instead of baselines | decentralized matrix sensing · Matrix sensing smallest Hessian eigenvalue, d = 50, ring (random) | smallest Hessian eigenvalue: -0.0329 -> −1.78e−5 | Improved |
| use PEDESTAL instead of baselines | decentralized matrix sensing · Matrix sensing smallest Hessian eigenvalue, d = 100, exponential (Dirichlet) | smallest Hessian eigenvalue: -0.0186 -> −3.63e−6 | Improved |
| use PEDESTAL instead of baselines | decentralized matrix factorization · MovieLens-100k, ring/toroidal/exponential, random and Dirichlet | loss function value | Improved |
| variance-reduced methods (D-GET, D-SPIDER-SFO) versus SGD-based D-PSGD and GTDSGD | baselines · MovieLens-100k matrix factorization | escaping saddle point / loss | Hurt |
| reduced gradient noise from variance reduction | decentralized baselines · Matrix factorization | saddle escape ability | Hurt |
| change data distribution from random to Dirichlet Dir20(0.3) | PEDESTAL · Matrix sensing, three topologies | saddle escape | No clear effect |
| vary network topology (ring, toroidal, exponential graph) | PEDESTAL · Matrix sensing d=50, d=100 | smallest Hessian eigenvalue: −1.78e−5 / −4.18e−5 / −1.09e−6 | No clear effect |
| increase dimension from d=50 to d=100 | all methods · Matrix sensing | smallest Hessian eigenvalue of baselines: -0.0332 -> -0.0184 | Mixed |
Closest papers by meaning
- An Online Optimization Perspective on First-Order and Zero-Order Decentralized Nonsmooth Nonconvex Stochastic Optimization (ICML 2024 · similarity 0.91)
- Decentralized Gradient-Free Methods for Stochastic Non-smooth Non-convex Optimization (AAAI 2024 · similarity 0.90)
- Decentralized Convex Finite-Sum Optimization with Better Dependence on Condition Numbers (ICML 2024 · similarity 0.90)
- Second-Order Convergence in Private Stochastic Non-Convex Optimization (NeurIPS 2025 · similarity 0.88)
- Decentralized Sum-of-Nonconvex Optimization (AAAI 2024 · similarity 0.88)
- Decentralized Projected Riemannian Stochastic Recursive Momentum Method for Nonconvex Optimization (AAAI 2025 · similarity 0.88)
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.
