Langevin algorithm #machine_learning #probability Overview Suppose function f:ℝd→ℝf: \mathbb{R}^d \to \mathbb{R}, with access to gradient ∇f\nabla f, where ∫exp(−f(x))dx\int \exp(-f(x))\, dx finite, #incomplete See also Langevin equation Brownian motion gradient descent References https://www.chrismusco.com/rdad2025/notes/faster_llms_annotated.pdf https://fa.bianp.net/blog/2023/ulaq/#sec1 https://en.wikipedia.org/wiki/Metropolis-adjusted_Langevin_algorithm