Research map / Optimization / Nonconvex and smooth optimization
On Convergence of Distributed Approximate Newton Methods: Globalization, Sharper Bounds and Beyond
Xiao-Tong Yuan, Ping Li
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 |
|---|---|---|---|
| add backtracking line search to DANE (DANE-LS) instead of plain InexactDANE | InexactDANE · synthetic logistic regression (p=200, m in {4,16,32}) | convergence stability | Improved |
| add heavy-ball momentum to DANE (DANE-HB, DANE-HB-LM) versus DANE-LS and InexactDANE | DANE-LS · synthetic logistic regression (p=200, m in {4,16,32}) | communication efficiency | Improved |
| add line search on rcv1.binary logistic regression | InexactDANE · rcv1.binary (p=47236, N=20242) | convergence behavior | No clear effect |
| add line search on gisette logistic regression | InexactDANE and AIDE · gisette (p=5000, N=6000) | convergence stability | Improved |
| increase number of machines m for DANE-LS (ablation) | DANE-LS · synthetic ridge regression (p in {200,500}, N=10p) | communication rounds to reach 1e-6: scales roughly linearly with sqrt(m) | Hurt |
| increase number of machines m for DANE-HB and DANE-HB-LM (ablation) | DANE-HB · synthetic ridge and logistic regression (p in {200,500}, N=10p) | communication rounds to reach 1e-6: scales roughly linearly with m^(1/4) | Hurt |
| decrease regularization parameter gamma (ablation) | DANE-LS, DANE-HB, DANE-HB-LM · synthetic regression (p in {200,500}) | communication rounds | Improved |
| compare DANE-HB-LM against DiSCO, DSVRG, GIANT on gisette | DiSCO · gisette (p=5000, N=6000) | communication efficiency: DANE-HB-LM > DiSCO >= DANE-HB | Improved |
| compare DANE-HB against DSVRG on rcv1.binary | DSVRG · rcv1.binary (p=47236, N=20242) | communication efficiency: DANE-HB >= DANE-HB-LM >= DANE-LS > DSVRG | Improved |
Closest papers by meaning
- Approximate Newton Methods and Their Local Convergence (ICML 2017 · similarity 0.89)
- Communication-Efficient Distributed Optimization in Networks with Gradient Tracking and Variance Reduction (JMLR 2020 · similarity 0.89)
- Federated Optimization with Doubly Regularized Drift Correction (ICML 2024 · similarity 0.87)
- Stabilized Proximal-Point Methods for Federated Optimization (NeurIPS 2024 · similarity 0.86)
- Accelerating Sinkhorn algorithm with sparse Newton iterations (ICLR 2024 · similarity 0.85)
- Convergence Analysis of Newton's Method for Neural Networks in the Overparameterized Limit (NeurIPS 2026 · similarity 0.85)
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.
