PulseAugur
EN
LIVE 07:49:45

New method estimates Fisher score for intractable likelihood maximization

Researchers have developed a novel gradient-based optimization method for likelihood maximization when the likelihood function is intractable but model simulations are available. This technique directly models the Fisher score using a local score matching approach with simulations from a localized parameter region. By using a linear parameterization for the surrogate score model, the method achieves a closed-form, least-squares solution, offering an efficient approximation to the Fisher score that smooths the likelihood objective and handles complex landscapes. Theoretical guarantees on the score estimator's bias are provided, and empirical results show superior performance against existing benchmarks on synthetic and real-world problems. AI

RANK_REASON This is a research paper detailing a new statistical method for likelihood maximization. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv stat.ML →

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

COVERAGE [1]

  1. arXiv stat.ML TIER_1 English(EN) · Sherman Khoo, Yakun Wang, Song Liu, Mark Beaumont ·

    Direct Fisher Score Estimation for Likelihood Maximization

    arXiv:2506.06542v2 Announce Type: replace Abstract: We study the problem of likelihood maximization when the likelihood function is intractable but model simulations are readily available. We propose a sequential, gradient-based optimization method that directly models the Fisher…