Researchers have established near-linear accuracy bounds for the Moreau--Yosida unadjusted Langevin algorithm (MYULA). The analysis provides an explicit step-size condition under which the invariant-measure bias relative to the Moreau-smoothed target is bounded by \(\\widetilde O(h)\\). This work combines the Moreau approximation bias with Wasserstein contraction to determine that \(\\widetilde O(\\varepsilon^{-1})\\) iterations are sufficient to achieve a desired accuracy for the Nth-iterate law. The stationary error is bounded directly, without requiring third derivatives or a Lipschitz Hessian, utilizing a Poisson-based estimate to convert a second-order stationary residual into a Wasserstein bound. AI
IMPACT Establishes theoretical bounds for sampling algorithms, potentially improving efficiency in machine learning model training.
RANK_REASON This is a research paper detailing theoretical advancements in sampling algorithms. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- DagsHub
- Hugging Face
- Langevin
- Mjurran
- Moreau
- Moreau--Yosida unadjusted Langevin algorithm
- Poisson
- Wasserstein
- Yoshida
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →