New algorithm achieves optimal regret in agnostic smoothed online learning

Researchers have developed a new algorithm for smoothed online learning that achieves sublinear regret in the agnostic setting without needing to sample the base measure or rely on perfectly predicted labels. This novel approach, named Gaussian Follow-The-Perturbed-Leader, is parameter-free, meaning it does not require knowledge of the base measure, smoothing parameter, or horizon. It achieves an optimal regret bound of \ufffdO(d\sqrt{T/\sigma}) for binary classes with VC dimension d, using a single call to an empirical risk minimization oracle per round. AI

IMPACT Introduces a more flexible and efficient method for online learning, potentially broadening its applicability in complex data scenarios.

RANK_REASON Academic paper detailing a new algorithm for online learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Gaussian Follow-The-Perturbed-Leader
• Gotit.pub
• Hugging Face
• IArxiv Recommender
• ScienceCast

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →