tfp.mcmc: Modern Markov Chain Monte Carlo Tools Built for Modern Hardware

在现代硬件和软件上运行马尔可夫链蒙特卡洛

Running Markov Chain Monte Carlo on Modern Hardware and Software

本章探讨利用GPU、TPU等现代硬件及PyTorch、JAX等软件加速马尔可夫链蒙特卡洛（MCMC）应用，提供加速模式、直觉解释、代码示例及潜在陷阱讨论，旨在提升MCMC计算效率。

在序列长度上并行化MCMC

Parallelizing MCMC Across the Sequence Length

本文提出在序列长度上并行化MCMC的方法，通过并行牛顿方法加速采样，提升效率。

Focusing on Difficult Directions for Learning HMC Trajectory Lengths

Hamiltonian Monte Carlo (HMC) is a premier Markov Chain Monte Carlo (MCMC) algorithm for continuous target distributions. Its full potential can only be unleashed when its problem-dependent hyperparameters are tuned well. The adaptation of one such hyperparameter, trajectory length ($τ$), has been closely examined by many research programs with the No-U-Turn Sampler (NUTS) coming out as the preferred method in 2011. A decade later, the evolving hardware profile has lead to the proliferation of personal and cloud based SIMD hardware in the form of Graphics and Tensor Processing Units (GPUs, TPU…

马尔可夫链蒙特卡洛应该运行多少次迭代？

For how many iterations should we run Markov chain Monte Carlo?

本文探讨MCMC运行迭代次数，回顾预热与采样阶段长度检查策略，在CPU/GPU背景下分析三个控制参数，围绕潜在变量推断、期望值分位数计算及估计量可靠性诊断展开，给出经验建议与理论依据。

gemlib.mcmc: 用于Metropolis-within-Gibbs采样方案的可组合内核

gemlib.mcmc: composable kernels for Metropolis-within-Gibbs sampling schemes

本文提出gemlib.mcmc框架，通过可组合内核抽象简化MCMC实现，利用范畴论中的writer monads实现内核组合，降低开发难度，使研究人员能高效应用先进算法。

Scalable Inference for Markov Processes with Intractable Likelihoods

Bayesian inference for Markov processes has become increasingly relevant in recent years. Problems of this type often have intractable likelihoods and prior knowledge about model rate parameters is often poor. Markov Chain Monte Carlo (MCMC) techniques can lead to exact inference in such models but in practice can suffer performance issues including long burn-in periods and poor mixing. On the other hand approximate Bayesian computation techniques can allow rapid exploration of a large parameter space but yield only approximate posterior distributions. Here we consider the combined use of appr…