A new framework called adversarially batched online learning has been introduced, where a learner's updates are restricted to specific time steps, effectively dividing the learning horizon into batches. This model is analyzed across four fundamental problems: stochastic bandits, online convex optimization, prediction with expert advice, and adversarial bandits. For stochastic bandits, regret bounds are established based on a complexity measure related to batch growth, showing that the penalty can increase with batch hardness. For the other three problems, the optimal regret is determined by the 2-norm of the batch-length sequence, achievable even by learners unaware of the batch structure. AI
IMPACT Introduces a new theoretical framework for online learning that could impact algorithm design in various machine learning applications.
RANK_REASON The cluster describes a new theoretical framework and analysis presented in an academic paper on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →