New SLICE method enables valid confidence sequences for simulated scientific models

A new paper introduces SLICE, a method for sequential likelihood-free inference that provides exact anytime-valid confidence sequences for scientific model parameters. This approach is particularly useful for models that can be simulated but have intractable likelihoods, a common challenge in frequentist uncertainty quantification. The method constructs confidence sequences by conformalizing scores into conditional e-values, which remain valid even under continuous monitoring, outperforming fixed-sample methods in such scenarios. AI

IMPACT Enables more robust uncertainty quantification for complex simulated models, potentially impacting scientific research and AI model evaluation.

RANK_REASON The cluster contains a new academic paper detailing a novel statistical methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• arXiv
• Conformal E-Values
• Hugging Face
• SLICE
• Stefano Cortinovis

• paper
• other

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