This paper introduces new algorithms for parallel stochastic convex optimization (SCO) that extend previous work beyond the standard Euclidean ($\ell_2$-norm) setting to $\ell_p$ geometries for $2 \le p < \infty$. The proposed ODC$_p$ algorithms, combined with multilevel Monte Carlo methods, establish new rates for optimization problems with $\ell_p$-Lipschitz objectives and composite minimization. Furthermore, by integrating with a novel $\ell_p$-generalized Gaussian-based stochastic gradient estimator called $p$-GREG, the paper presents two parallel algorithms, EpochDA and $\ell_p$-AC-SA, which achieve improved query depths compared to existing sequential methods, particularly in heavy-tailed noise models. AI
IMPACT Introduces novel optimization techniques that could improve the efficiency of training large-scale machine learning models.
RANK_REASON The cluster contains a research paper detailing new algorithms for a specific optimization problem. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- \ell_2$-norm
- \ell_p$
- \ell_p$-AC-SA
- \ell_p$ Geometries
- EpochDA
- Hugging Face Daily Papers
- Multilevel Monte Carlo method
- ODC$_p$
- $p$-GREG
- Stochastic Convex Optimization with Multiple Objectives
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →