Research map / Optimization / Nonconvex and smooth optimization
Replica Exchange for Non-Convex Optimization
Jing Dong, Xin T. Tong
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 exchange between GD and LD (GDxLD) instead of running LD alone with small temperature | LD · 2D Gaussian mixture (offline) | iterations to global minimum: 1.2 x 10^5 -> less than 300 | Improved |
| add exchange between SGD and SGLD (SGDxSGLD) instead of SGLD alone | SGLD · 2D Gaussian mixture (online KDE) | iterations to global minimum: 2 x 10^4 -> 10^3 | Improved |
| add exchange with LD to GD, avoiding local minima trapping | GD · 2D Gaussian mixture (offline) | convergence to global minimum | Improved |
| use non-swapping variant nGDxLD instead of swapping GDxLD (ablation) | GDxLD · 2D Gaussian mixture (offline) | convergence speed: less than 300 -> less than 400 iterations | No clear effect |
| increase LD temperature gamma to 1 in LD alone (ablation) | LD · 2D Gaussian mixture (offline) | convergence to global minimum | Hurt |
| vary temperature gamma from 0.5 to 10 (ablation) | GDxLD · 2D Gaussian mixture (offline) | convergence speed: 0.5 <= gamma <= 10 | No clear effect |
| vary step size h in 0.05 to 1 (ablation) | GDxLD · 2D Gaussian mixture (offline) | convergence speed: 0.05 <= h <= 1 | No clear effect |
| increase step size h to 1.5 (ablation) | GDxLD · 2D Gaussian mixture (offline) | convergence accuracy | Hurt |
| increase dimension of Rastrigin from 2 to 5 (ablation) | GDxLD · Rastrigin | iterations to global minimum: 5 x 10^2 -> 7 x 10^4 | Hurt |
| increase Rastrigin dimension to d>=10 (ablation) | GDxLD · Rastrigin | finding global minimum within 10^5 iterations: not found within 10^5 iterations | Hurt |
Closest papers by meaning
- ACCELERATING NONCONVEX LEARNING VIA REPLICA EXCHANGE LANGEVIN DIFFUSION (ICLR 2019 · similarity 0.92)
- Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics (ICML 2024 · similarity 0.90)
- Generalization of noisy SGD in unbounded non-convex settings (ICML 2025 · similarity 0.88)
- Symmetric Mean-field Langevin Dynamics for Distributional Minimax Problems (ICLR 2024 · similarity 0.87)
- The Expected Loss of Preconditioned Langevin Dynamics Reveals the Hessian Rank (AAAI 2024 · similarity 0.87)
- Regularized Langevin Dynamics for Combinatorial Optimization (ICML 2025 · 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.
