This paper introduces a novel approach to hypothesis testing in a density-free setting, where only simulated data from two probability distributions is available. The authors construct an e-test martingale to develop a sequential test that offers anytime-valid type-I error guarantees. This method also provides approximate growth optimality and geometrically decaying type-II error bounds, ultimately achieving asymptotic power one. The paper's contribution is a concise and effective solution for simulation-based sequential hypothesis testing without requiring analytic densities. AI
IMPACT Introduces a novel statistical method for analyzing simulated data, potentially impacting AI research that relies on hypothesis testing with limited or no analytic density information.
RANK_REASON The item is a research paper detailing a new methodology for hypothesis testing. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
