New DART method accelerates Markov chain mixing without gradients

Researchers have developed a novel method called Delayed Acceptance with Regularisation and Tempering (DART) to accelerate Markov chain mixing times without requiring gradient calculations. This approach leverages a localization principle to incorporate geometric information from surrogate densities, achieving an O(κ max{κ, d}) mixing time for strongly log-concave targets in d dimensions. DART's performance matches or surpasses existing methods like MALA, particularly in high-dimensional scenarios, and it has been demonstrated on complex models such as hierarchical spatial generalized linear mixed models. AI

IMPACT This research could lead to more efficient sampling methods for complex models, potentially impacting AI research that relies on probabilistic inference.

RANK_REASON Academic paper detailing a new statistical method. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv stat.ML →

• arXiv
• Delayed Acceptance with Regularisation and Tempering
• Gaussian Processes
• Langevin
• MALA
• Markov chain

• paper
• other

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