New Bayesian Inference Framework for Non-Stationary Bandits

Researchers have developed a new Bayesian inference framework called Weighted Sequential Bayesian (WSB) inference for non-stationary linear contextual bandits. This approach establishes a sequence of posteriors over reward parameters, allowing for a more principled randomized exploration strategy. The WSB framework has been instantiated into three new algorithms: WSB-LinUCB, WSB-RandLinUCB, and WSB-LinTS, which demonstrate competitive or superior performance compared to existing WRLS-based methods. AI

IMPACT Introduces a novel Bayesian approach to improve exploration strategies in non-stationary bandit problems, potentially leading to more efficient learning algorithms.

RANK_REASON The cluster contains an academic paper detailing a new statistical inference method for a machine learning problem. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• Abdullah Akgül
• Non-Stationary Linear Contextual Bandits
• Weighted Regularized Least-Squares
• Weighted Sequential Bayesian Inference
• WSB-LinTS
• WSB-LinUCB
• WSB-RandLinUCB

• paper
• other

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