Research map / Optimization / Nonconvex and smooth optimization
Mirrored Langevin Dynamics
Ya-Ping Hsieh, Ali Kavis, Paul Rolland, Volkan Cevher
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 MLD with entropic mirror map instead of SGRLD for sampling an 11-dimensional sparse Dirichlet posterior | Mirrored Langevin Dynamics · Synthetic 11-dim Dirichlet posterior, first dimension | total variation error | Improved |
| Use MLD instead of SGRLD, looking at the other dimensions with fewer observations | Mirrored Langevin Dynamics · Synthetic 11-dim Dirichlet posterior, other dimensions | total variation error | Improved |
| Use SMLD-approximate instead of SGRLD for LDA | SMLD-approximate · LDA on Wikipedia corpus, 100,000 documents | test perplexity | Improved |
| Replace exp(y) by the linear approximation max{0, 1+y} in the stochastic gradient | SMLD · LDA on Wikipedia corpus | numerical stability | Improved |
| Use higher-order Taylor approximation or a small threshold max{eps, 1+y} instead of the linear approximation | SMLD-approximate · LDA on Wikipedia corpus | test perplexity | No clear effect |
| Use the dual-space mini-batch gradient instead of the primal mini-batch gradient | Stochastic Mirrored Langevin Dynamics · Theory (Lemma 2) | stochastic gradient variance | Improved |
| Apply mirror map to a strongly log-concave constrained target instead of existing MYULA/PLMC schemes | MLD · Strongly log-concave target with bounded domain | convergence rate | Improved |
| Use final iterate rather than random iterate in SMLD | SMLD · Practice | performance | Improved |
Closest papers by meaning
- Efficient constrained sampling via the mirror-Langevin algorithm (NeurIPS 2021 · similarity 0.94)
- Fast Non-Log-Concave Sampling under Nonconvex Equality and Inequality Constraints with Landing (NeurIPS 2025 · similarity 0.89)
- Constrained Sampling with Primal-Dual Langevin Monte Carlo (NeurIPS 2024 · similarity 0.89)
- A connection between Tempering and Entropic Mirror Descent (ICML 2024 · similarity 0.89)
- Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics (ICML 2024 · similarity 0.88)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 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.
