General Construction of Irreversible Kernel in Markov Chain Monte Carlo

Similar documents
arxiv: v2 [cond-mat.stat-mech] 13 Oct 2010

Irreversible Markov-Chain Monte Carlo methods

Pattern Recognition and Machine Learning. Bishop Chapter 11: Sampling Methods

Introduction to Machine Learning CMU-10701

Computational statistics

Markov Chain Monte Carlo Using the Ratio-of-Uniforms Transformation. Luke Tierney Department of Statistics & Actuarial Science University of Iowa

Markov Chain Monte Carlo

A Search and Jump Algorithm for Markov Chain Monte Carlo Sampling. Christopher Jennison. Adriana Ibrahim. Seminar at University of Kuwait

Stat 535 C - Statistical Computing & Monte Carlo Methods. Lecture February Arnaud Doucet

Session 3A: Markov chain Monte Carlo (MCMC)

Markov chain Monte Carlo

Markov Chain Monte Carlo (MCMC)

Convex Optimization CMU-10725

A = {(x, u) : 0 u f(x)},

Their Statistical Analvsis. With Web-Based Fortran Code. Berg

Monte Carlo simulation inspired by computational optimization. Colin Fox Al Parker, John Bardsley MCQMC Feb 2012, Sydney

Markov Chain Monte Carlo Simulations and Their Statistical Analysis An Overview

ST 740: Markov Chain Monte Carlo

CSC 2541: Bayesian Methods for Machine Learning

University of Toronto Department of Statistics

17 : Markov Chain Monte Carlo

A quick introduction to Markov chains and Markov chain Monte Carlo (revised version)

Computer Vision Group Prof. Daniel Cremers. 11. Sampling Methods: Markov Chain Monte Carlo

Minicourse on: Markov Chain Monte Carlo: Simulation Techniques in Statistics

Bayesian Methods for Machine Learning

SAMPLING ALGORITHMS. In general. Inference in Bayesian models

Markov Processes. Stochastic process. Markov process

April 20th, Advanced Topics in Machine Learning California Institute of Technology. Markov Chain Monte Carlo for Machine Learning

Markov Chain Monte Carlo The Metropolis-Hastings Algorithm

Reminder of some Markov Chain properties:

STAT 425: Introduction to Bayesian Analysis

Hastings-within-Gibbs Algorithm: Introduction and Application on Hierarchical Model

Introduction to Bayesian methods in inverse problems

Inference in state-space models with multiple paths from conditional SMC

Simulated Annealing for Constrained Global Optimization

Markov chain Monte Carlo

Deblurring Jupiter (sampling in GLIP faster than regularized inversion) Colin Fox Richard A. Norton, J.

Rare Event Sampling using Multicanonical Monte Carlo

Markov Chain Monte Carlo (MCMC) and Model Evaluation. August 15, 2017

Markov Chain Monte Carlo Method

Monte Carlo Methods. Leon Gu CSD, CMU

Brief introduction to Markov Chain Monte Carlo

Monte Carlo methods for sampling-based Stochastic Optimization

Probabilistic Graphical Models Lecture 17: Markov chain Monte Carlo

Bayesian Phylogenetics:

Example: Ground Motion Attenuation

MCMC Sampling for Bayesian Inference using L1-type Priors

16 : Markov Chain Monte Carlo (MCMC)

Markov Chain Monte Carlo Inference. Siamak Ravanbakhsh Winter 2018

18 : Advanced topics in MCMC. 1 Gibbs Sampling (Continued from the last lecture)

Bayesian Inference and MCMC

Pseudo-marginal MCMC methods for inference in latent variable models

Classical boson sampling algorithms and the outlook for experimental boson sampling

Likelihood Inference for Lattice Spatial Processes

F denotes cumulative density. denotes probability density function; (.)

Monte Carlo Markov Chains: A Brief Introduction and Implementation. Jennifer Helsby Astro 321

CSC 446 Notes: Lecture 13

Computer Vision Group Prof. Daniel Cremers. 11. Sampling Methods

MONTE CARLO METHODS. Hedibert Freitas Lopes

CS242: Probabilistic Graphical Models Lecture 7B: Markov Chain Monte Carlo & Gibbs Sampling

(5) Multi-parameter models - Gibbs sampling. ST440/540: Applied Bayesian Analysis

Metropolis Hastings. Rebecca C. Steorts Bayesian Methods and Modern Statistics: STA 360/601. Module 9

Forward Problems and their Inverse Solutions

Quantifying Uncertainty

Markov Chains and MCMC

Introduction to Computational Biology Lecture # 14: MCMC - Markov Chain Monte Carlo

The Bias-Variance dilemma of the Monte Carlo. method. Technion - Israel Institute of Technology, Technion City, Haifa 32000, Israel

arxiv: v1 [stat.co] 18 Feb 2012

Advanced Statistical Computing

Improving the Asymptotic Performance of Markov Chain Monte-Carlo by Inserting Vortices

Advances and Applications in Perfect Sampling

Principles of Bayesian Inference

6 Markov Chain Monte Carlo (MCMC)

Theory of Stochastic Processes 8. Markov chain Monte Carlo

MCMC and Gibbs Sampling. Sargur Srihari

Reducing The Computational Cost of Bayesian Indoor Positioning Systems

Control Variates for Markov Chain Monte Carlo

Sparse Sensing for Statistical Inference

Lecture 7 and 8: Markov Chain Monte Carlo

Time-Sensitive Dirichlet Process Mixture Models

Convergence Rate of Markov Chains

The simple slice sampler is a specialised type of MCMC auxiliary variable method (Swendsen and Wang, 1987; Edwards and Sokal, 1988; Besag and Green, 1

Sample-based UQ via Computational Linear Algebra (not MCMC) Colin Fox Al Parker, John Bardsley March 2012

Computer Vision Group Prof. Daniel Cremers. 10a. Markov Chain Monte Carlo

Markov Chain Monte Carlo methods

Introduction to Machine Learning CMU-10701

The XY-Model. David-Alexander Robinson Sch th January 2012

Adaptive Monte Carlo methods

Polynomial accelerated MCMC... and other sampling algorithms inspired by computational optimization

CPSC 540: Machine Learning

Stochastic Simulation

Markov chain Monte Carlo Lecture 9

Markov chain Monte Carlo: Some practical implications. of theoretical results

Monte Carlo in Bayesian Statistics

Markov Chain Monte Carlo methods

Sampling Rejection Sampling Importance Sampling Markov Chain Monte Carlo. Sampling Methods. Oliver Schulte - CMPT 419/726. Bishop PRML Ch.

Hamiltonian Monte Carlo for Scalable Deep Learning

STA 4273H: Statistical Machine Learning

Critical Dynamics of Two-Replica Cluster Algorithms

A generalization of the Multiple-try Metropolis algorithm for Bayesian estimation and model selection

Transcription:

General Construction of Irreversible Kernel in Markov Chain Monte Carlo Metropolis heat bath Suwa Todo Department of Applied Physics, The University of Tokyo Department of Physics, Boston University (from April) Collaborator : Synge Todo 1

Markov Chain Monte Carlo Metropolis et al. (1953) Monte Carlo method Numerical integration method using a (quasi-)random number Uniformly random sampling suffers from curse of dimensionality Importance sampling P Markov chain Monte Carlo (MCMC) = Numerical method generating states from a target distribution Γ Markov chain Sample correlation Autocorrelation time Needed rough number of steps for forgetting the past C(t) = A i+ta i A 2 A 2 A 2 2

For Efficient Sampling Three key points for reducing the autocorrelation in MCMC 1. Determination of ensemble : Multicanonical, Exchange Monte Carlo 2. Selection of candidate states: Cluster algorithm, Hybrid Monte Carlo 3. Optimization of transition kernel: Metropolis, Heat bath (Gibbs sampler) Let us consider a discrete variable Heat bath algorithm (Gibbs sampler) Barker (1965), Geman and Geman (1984) Metropolis (Metropolis-Hastings) algorithm Metropolis et al. (1953), Hastings (197) p(c i c j )= w(c j) k w(c k) p(c i c j )= 1 n 1 min 1, w(c j) w(c i ) The above algorithms are the standard methods in MCMC. However, they are not optimal! 3

What is Optimal? P MG = P IMG = π 2 1 π 1 1... π 3 1 π 1 π 3.. π n 1 π 1 π 1 1 π 1 π 1 1 π 1 π 2 1 π 2 π 1 1 π 1 π 2 1 π 2 π 3 1 π 3 1 π 2 1............ π n 1 π 2 π n 1 π 3 1... y 1 y 1 y 1 w 2 w 1 y 1 y 2 y 2 w 3 w 1 y 1 w 3. w n w 1 y 1 w 2 y 2 y 3......... w n w 2 y 2 w n w 3 y 3 1 y 1 y 2... Peskun (1973) Metropolized Gibbs Sampler Liu (1996) = Gibbs sampler excluding the current state + Metropolis, Pij MG = min(π i /(1 π j ),π i /(1 π i )) Iterative Metropolized Gibbs Sampler, π i = w i j w j π 1 π 2 π n Frigessi et al. (1992) 4

Reversibility of Markov Chain 5

Stochastic Flow 6

Geometric Allocation : n=2 7

Inevitable Rejection? : n=4 σ i 8

H.S. and Todo, (21) Phys. Rev. Lett., 15, 1263 New Algorithm w 1 9

Comparison with Conventional Methods Metropolis heat bath Algorithm 1 1

Acceleration in Potts Model Square lattice Autocorrelation time of the structure factor 1 3 q = 4 q = 8 Binning analysis Ferromagnetic q-state Potts model σ i Wu (1982) int 1 2 1 1 Metropolis Heat Bath Metropolized Gibbs Iterative MG Optimal Average Optimal Rejection Hwang (25) 1.86.9.94.98 Metropolis Heat bath (Gibbs sampler) T.74.76.78 Phase transition temperature Continuous 1st order Significantly speed up! 11

Dynamical Exponent 3.5 3 Autocorrelation time of the structure factor log 1 int 2.5 2 1.5 1.5 -.5.6.8 1 1.2 1.4 1.6 1.8 2 log 1 L Metropolis Heat Bath Metropolized Gibbs Iterative MG Optimal Average Optimal Rejection 2.3 * x - 1.8 τ int L z z 2.3 Same dynamical exponent. But we always gain a factor (6 than Metropolis). 12

Convergence Speed Up Two criteria of Markov chain 1. Short autocorrelation time = small asymptotic variance = high-sampling efficiency 2. Rapid distribution convergence = short burn-in (thermalization) process Square lattice L = 32 Relaxation of the order parameter Start with all up state log 1 m -.5 -.1 -.15 -.2 -.25 -.3 Metropolis Heat Bath Metropolized Gibbs Iterative MG Optimal Average Optimal Rejection -.35.5 1 1.5 2 2.5 3 log 1 t Our algorithm is the best! Not only the sampling efficiency but also the distribution convergence is improved. 13

Extension to Continuous Variable 1.5 1 5-2 2-2 2-5 -1-1 -5 5 1-1 -5 5 1 14

Comparison with Conventional Methods 1 1e+6 1 Heat Bath Overrelaxation Ordered Overrelaxation present c=.4, w=.5 int 1 1 1 1.1 1 1 1 1 2 / 1 15

Beyond Metropolis Frenkel et al. (21), Liu et al. (2), Qin et al. (21) Neal (1994) 16

Decreasing Rejection Rate P (x 1, x 2 ) exp (x1 x 2 ) 2 2σ1 2 + (x 1 + x 2 ) 2 2σ 2 2 (x1 x 2 ) 2 2σ 2 1 + (x 1 + x 2 ) 2 2σ 2 2 h + h2 4. 5 5.6.5.4.3.2.1 N(, 1) Metropolis present n=3 present n=4 2 4 6 8 1 12 14 16 18 2 σ 2 /σ 1 35 3 25 2 15 1 5 Metropolis present n=3 present n=4 5 5 τ int of (x 1 +x 2 ) 2 2 4 6 8 1 12 14 16 18 2 1 5 σ 2 /σ 1 17

cf. H.S. and Todo, Phys. Rev. Lett., 15, 1263 (21) arxiv:116.3562 Summary Metropolis heat bath Suwa Todo 18