Research map / Optimization / Distributed and compressed
Decentralized Sparse Linear Regression via Gradient-Tracking
Marie Maros, Gesualdo Scutari, Ying Sun, Guang Cheng
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 |
|---|---|---|---|
| increase alpha = s log d / N from 0.04 to 1 (fewer local samples) (ablation) | DGT · Synthetic sparse regression, d=20000, m=50 | convergence rate and statistical precision: alpha 0.04, 0.2, 1 (n = 25, 5, 1) | Hurt |
| increase dimension d with alpha fixed at about 0.2 (ablation) | DGT · Synthetic sparse regression, m=50 ER graph | convergence rate: d = 5000, 10000, 20000 | No clear effect |
| increase network size m while keeping rho about 0.18 (ablation) | DGT · Synthetic, d=5000, N=2500, ER graphs | convergence rate and estimation error: m = 50, 625, 1250, 2500 with rho about 0.18 | Hurt |
| grow network size with fixed link probability (rho = O(1/sqrt(m))) (ablation) | DGT · Synthetic, d=5000, ER graphs p=0.87 | convergence under growing m: rho 0.167 -> 0.031 as m 50 -> 2500 | No clear effect |
| use line graph instead of Erdos-Renyi graph (ablation) | DGT · Communication rounds per iteration, m=2500 | communication rounds: 19 -> 118911225 | Hurt |
| use DGT or DGD2 instead of DGD-ATC | decentralized sparse regression · Communication rounds vs dimension d (m=120) | communication rounds | Improved |
| use DGT instead of DGD-CTA on fully connected network | decentralized sparse regression · Communication rounds to centralized precision (m=120) | communication rounds | Improved |
| use DGT instead of centralized PGD | Communities and Crime, m=41 · Communities and Crime | train and test error | No clear effect |
| add multiple communication rounds per gradient (3 rounds) (ablation) | DGT · Communities and Crime, m=41 | effective connectivity rho: 0.5236 -> 0.1435 | Improved |
Closest papers by meaning
- Acceleration in Distributed Sparse Regression (NeurIPS 2022 · similarity 0.92)
- DGD^2: A Linearly Convergent Distributed Algorithm For High-dimensional Statistical Recovery (NeurIPS 2022 · similarity 0.91)
- Implicit Regularization of Decentralized Gradient Descent for Sparse Regression (NeurIPS 2024 · similarity 0.91)
- An Improved Analysis of Gradient Tracking for Decentralized Machine Learning (NeurIPS 2021 · similarity 0.90)
- Double Stochasticity Gazes Faster: Snap-Shot Decentralized Stochastic Gradient Tracking Methods (ICML 2024 · similarity 0.87)
- Achieving Linear Convergence with Parameter-Free Algorithms in Decentralized Optimization (NeurIPS 2024 · similarity 0.86)
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.
