DOCSLIB.ORG

Hamiltonian Monte Carlo Without Detailed Balance

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Hamiltonian Monte Carlo Without Detailed Balance Hamiltonian Monte Carlo Without Detailed Balance Jascha Sohl-Dickstein [email protected] Stanford University, Palo Alto. Khan Academy, Mountain View Mayur Mudigonda [email protected] Redwood Institute for Theoretical Neuroscience, University of California at Berkeley Michael R. DeWeese [email protected] Redwood Institute for Theoretical Neuroscience, University of California at Berkeley Abstract rithm converges to the correct distribution is via a concept We present a method for performing Hamiltonian known as detailed balance. Sampling algorithms based on Monte Carlo that largely eliminates sample re- detailed balance are powerful because they allow samples jection. In situations that would normally lead to from any target distribution to be generated from almost rejection, instead a longer trajectory is computed any proposal distribution, using for instance Metropolis- until a new state is reached that can be accepted. Hastings acceptance criteria (Hastings, 1970). However, This is achieved using Markov chain transitions detailed balance also suffers from a critical flaw. Pre- that satisfy the fixed point equation, but do not cisely because the forward and reverse transitions occur satisfy detailed balance. The resulting algorithm with equal probability, detailed balance driven samplers go significantly suppresses the random walk behav- backwards exactly as often as they go forwards. The state ior and wasted function evaluations that are typ- space is thus explored via a random walk over distances ically the consequence of update rejection. We longer than those traversed by a single draw from the pro- posal distribution. A random walk only travels a distance demonstrate a greater than factor of two improve- 1 ment in mixing time on three test problems. We dN 2 in N steps, where d is the characteristic step length. release the source code as Python and MATLAB The current state-of-the-art sampling algorithm for proba- packages. bility distributions with continuous state spaces is Hamilto- nian Monte Carlo (HMC) (Duane et al., 1987; Neal, 2010). By extending the state space to include auxiliary momen- 1. Introduction tum variables, and then using Hamiltonian dynamics to tra- High dimensional and otherwise computationally expen- verse long iso-probability contours in this extended state sive probabilistic models are of increasing importance for space, HMC is able to move long distances in state space in such diverse tasks as modeling the folding of proteins a single update step. However, HMC still relies on detailed (Schutte¨ & Fischer, 1999), the structure of natural images balance to accept or reject steps, and as a result still behaves (Culpepper et al., 2011), or the activity of networks of neu- like a random walk – just a random walk with a longer step rons (Cadieu & Koepsell, 2010). length. Previous attempts to address this have combined multiple Markov steps that individually satisfy detailed bal- Sampling from the described distribution is typically the ance into a composite step that does not (Horowitz, 1991), bottleneck when working with these probabilistic models. with limited success (Kennedy & Pendleton, 1991). Sampling is commonly required when training a proba- bilistic model, when evaluating the model’s performance, The No-U-Turn Sampler (NUTS) sampling package (Hoff- when performing inference, and when taking expectations man & Gelman, 2011) and the windowed acceptance (MacKay, 2003). Therefore, work that improves sampling method of (Neal, 1994) both consider Markov transitions is fundamentally important. within a set of discrete states generated by repeatedly sim- ulating Hamiltonian dynamics. NUTS generates a set of The most common way to guarantee that a sampling algo- candidate states around the starting state by running Hamil- Proceedings of the 31 st International Conference on Machine tonian dynamics forwards and backwards until the trajec- Learning, Beijing, China, 2014. JMLR: W&CP volume 32. Copy- tory doubles back on itself, or a slice variable constraint is right 2014 by the author(s). violated. It then chooses a new state at uniform from the Hamiltonian Monte Carlo Without Detailed Balance candidate states. In windowed acceptance, a transition is fixed point equation (Equation2). Detailed balance guar- proposed between a window of states at the beginning and antees that if samples are drawn from the equilibrium dis- end of a trajectory, rather than the first state and last state. tribution p (x), then for every pair of states x and x0 the With the selected window, a single state is then chosen us- probability of transitioning from state x to state x0 is iden- ing Boltzmann weightings. Both NUTS and the windowed tical to that of transitioning from state x0 to x, acceptance method rely on detailed balance to choose the candidate state from the discrete set. p (x) T (x0jx) = p (x0) T (xjx0) : (3) Here we present a novel discrete representation of the HMC By substitution for T (x0jx) in the left side of Equation2, state space and transitions. Using this representation, we it can be seen that if Equation3 is satisfied, then the fixed derive a method for performing HMC while abandoning point equation is also satisfied. detailed balance altogether, by directly satisfying the fixed An appealing aspect of detailed balance is that a tran- point equation restricted to the discrete state space. As a sition distribution satisfying it can be easily constructed result, random walk behavior in the sampling algorithm is from nearly any proposal distribution, using Metropolis- greatly reduced, and the mixing rate of the sampler is sub- Hastings acceptance/rejection rules (Hastings, 1970). A stantially improved. primary drawback of detailed balance, and of Metropolis- Hastings, is that the resulting Markov chains always engage 2. Sampling in random walk behavior, since by definition detailed bal- ance depends on forward and reverse transitions happening We begin by briefly reviewing some key concepts related with equal probability. to sampling. The goal of a sampling algorithm is to draw characteristic samples x 2 RN from a target probability The primary advance in this paper is demonstrating how distribution p (x). Without loss of generality, we will as- HMC sampling can be performed without resorting to de- sume that p (x) is determined by an energy function E (x), tailed balance. 1 p (x) = exp (−E (x)) : (1) Z 3. Hamiltonian Monte Carlo 2.1. Markov Chain Monte Carlo Hamiltonian Monte Carlo (HMC) can traverse long dis- tances in state space with single Markov transitions. It Markov Chain Monte Carlo (MCMC) (Neal, 1993) is does this by extending the state space to include auxiliary commonly used to sample from probabilistic models. In momentum variables, and then simulating Hamiltonian dy- MCMC a chain of samples is generated by repeatedly namics to move long distances along iso-probability con- drawing new samples x0 from a conditional probability dis- tours in the expanded state space. tribution T (x0jx), where x is the previous sample. Since T (x jx) is a probability density over x , R T (x jx) dx = 0 0 0 0 3.1. Extended state space 1 and T (x0jx) ≥ 0. The state space is extended by the addition of momentum 2.2. Fixed Point Equation variables v 2 RN , with identity-covariance Gaussian dis- tribution, An MCMC algorithm must satisfy two conditions in or- der to generate samples from the target distribution p (x). N 1 T p (v) = (2π)− 2 exp − v v : (4) The first is mixing, which requires that repeated application 2 of T (x0jx) must eventually explore the full state space of p (x). The second condition is that the target distribution We refer to the combined state space of x and v as ζ, such f g p (x) must be a fixed point of T (x0jx). This second condi- that ζ = x; v . The corresponding joint distribution is tion can be expressed by the fixed point equation, N (2π)− 2 Z p (ζ) = p (x; v) = p (x) p (v) = exp (−H (ζ)) ; p (x) T (x0jx) dx = p (x0) ; (2) Z (5) which requires that when T (x0jx) acts on p (x), the result- 1 H (ζ) = H (x; v) = E (x) + vT v: (6) ing distribution is unchanged. 2 has the same form as total energy in a physical sys- 2.3. Detailed Balance H (ζ) tem, where E (x) is the potential energy for position x and 1 T Detailed balance is the most common way of guaranteeing 2 v v is the kinetic energy for momentum v (mass is set that the Markov transition distribution T (x0jx) satisfies the to one). Hamiltonian Monte Carlo Without Detailed Balance In HMC samples from p (x) are generated by drawing sam- ples from the joint distribution p (x; v), and using only the x variables as samples from the desired distribution. ⇣ 1 F⇣ L− F⇣ L⇣ 3.2. Hamiltonian dynamics L⇣ Hamiltonian dynamics govern how physical systems F⇣ ⇣ ⇣ evolve with time. It might be useful to imagine the trajec- tory of a skateboarder rolling in an empty swimming pool. 1 As she rolls downwards she exchanges potential energy for LF⇣ L− ⇣ kinetic energy, and the magnitude of her velocity increases. ⇣ As she rolls up again she exchanges kinetic energy back for R (β) ⇣ potential energy. In this fashion she is able to traverse long (a) (b) distances across the swimming pool, while at the same time maintaining constant total energy over her entire trajectory. Figure 1. (a) The action of operators involved in Hamiltonian Monte Carlo (HMC). The base of each red or green arrow repre- In HMC, we treat H (ζ) as the total energy of a physi- sents the position x, and the length and direction of each of these cal system, with spatial coordinate x, velocity v, potential arrows represents the momentum v. The flip operator F reverses 1 T energy E (x), and kinetic energy 2 v v.
Load more

Recommended publications
  • The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo
    Journal of Machine Learning Research 15 (2014) 1593-1623 Submitted 11/11; Revised 10/13; Published 4/14 The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo Matthew D. Hoffman [email protected] Adobe Research 601 Townsend St. San Francisco, CA 94110, USA Andrew Gelman [email protected] Departments of Statistics and Political Science Columbia University New York, NY 10027, USA Editor: Anthanasios Kottas Abstract Hamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) algorithm that avoids the random walk behavior and sensitivity to correlated parameters that plague many MCMC methods by taking a series of steps informed by first-order gradient information. These features allow it to converge to high-dimensional target distributions much more quickly than simpler methods such as random walk Metropolis or Gibbs sampling. However, HMC's performance is highly sensitive to two user-specified parameters: a step size and a desired number of steps L. In particular, if L is too small then the algorithm exhibits undesirable random walk behavior, while if L is too large the algorithm wastes computation. We introduce the No-U-Turn Sampler (NUTS), an extension to HMC that eliminates the need to set a number of steps L. NUTS uses a recursive algorithm to build a set of likely candidate points that spans a wide swath of the target distribution, stopping automatically when it starts to double back and retrace its steps. Empirically, NUTS performs at least as efficiently as (and sometimes more efficiently than) a well tuned standard HMC method, without requiring user intervention or costly tuning runs.
    [Show full text]
  • Arxiv:1801.02309V4 [Stat.ML] 11 Dec 2019
    Journal of Machine Learning Research 20 (2019) 1-42 Submitted 4/19; Revised 11/19; Published 12/19 Log-concave sampling: Metropolis-Hastings algorithms are fast Raaz Dwivedi∗;y [email protected] Yuansi Chen∗;} [email protected] Martin J. Wainwright};y;z [email protected] Bin Yu};y [email protected] Department of Statistics} Department of Electrical Engineering and Computer Sciencesy University of California, Berkeley Voleon Groupz, Berkeley Editor: Suvrit Sra Abstract We study the problem of sampling from a strongly log-concave density supported on Rd, and prove a non-asymptotic upper bound on the mixing time of the Metropolis-adjusted Langevin algorithm (MALA). The method draws samples by simulating a Markov chain obtained from the discretization of an appropriate Langevin diffusion, combined with an accept-reject step. Relative to known guarantees for the unadjusted Langevin algorithm (ULA), our bounds show that the use of an accept-reject step in MALA leads to an ex- ponentially improved dependence on the error-tolerance. Concretely, in order to obtain samples with TV error at most δ for a density with condition number κ, we show that MALA requires κd log(1/δ) steps from a warm start, as compared to the κ2d/δ2 steps establishedO in past work on ULA. We also demonstrate the gains of a modifiedO version of MALA over ULA for weakly log-concave densities. Furthermore, we derive mixing time bounds for the Metropolized random walk (MRW) and obtain (κ) mixing time slower than MALA. We provide numerical examples that support ourO theoretical findings, and demonstrate the benefits of Metropolis-Hastings adjustment for Langevin-type sampling algorithms.
    [Show full text]
  • An Introduction to Data Analysis Using the Pymc3 Probabilistic Programming Framework: a Case Study with Gaussian Mixture Modeling
    An introduction to data analysis using the PyMC3 probabilistic programming framework: A case study with Gaussian Mixture Modeling Shi Xian Liewa, Mohsen Afrasiabia, and Joseph L. Austerweila aDepartment of Psychology, University of Wisconsin-Madison, Madison, WI, USA August 28, 2019 Author Note Correspondence concerning this article should be addressed to: Shi Xian Liew, 1202 West Johnson Street, Madison, WI 53706. E-mail: [email protected]. This work was funded by a Vilas Life Cycle Professorship and the VCRGE at University of Wisconsin-Madison with funding from the WARF 1 RUNNING HEAD: Introduction to PyMC3 with Gaussian Mixture Models 2 Abstract Recent developments in modern probabilistic programming have offered users many practical tools of Bayesian data analysis. However, the adoption of such techniques by the general psychology community is still fairly limited. This tutorial aims to provide non-technicians with an accessible guide to PyMC3, a robust probabilistic programming language that allows for straightforward Bayesian data analysis. We focus on a series of increasingly complex Gaussian mixture models – building up from fitting basic univariate models to more complex multivariate models fit to real-world data. We also explore how PyMC3 can be configured to obtain significant increases in computational speed by taking advantage of a machine’s GPU, in addition to the conditions under which such acceleration can be expected. All example analyses are detailed with step-by-step instructions and corresponding Python code. Keywords: probabilistic programming; Bayesian data analysis; Markov chain Monte Carlo; computational modeling; Gaussian mixture modeling RUNNING HEAD: Introduction to PyMC3 with Gaussian Mixture Models 3 1 Introduction Over the last decade, there has been a shift in the norms of what counts as rigorous research in psychological science.
    [Show full text]
  • Hamiltonian Monte Carlo Swindles
    Hamiltonian Monte Carlo Swindles Dan Piponi Matthew D. Hoffman Pavel Sountsov Google Research Google Research Google Research Abstract become indistinguishable (Mangoubi and Smith, 2017; Bou-Rabee et al., 2018). These results were proven in Hamiltonian Monte Carlo (HMC) is a power- service of proofs of convergence and mixing speed for ful Markov chain Monte Carlo (MCMC) algo- HMC, but have been used by Heng and Jacob (2019) rithm for estimating expectations with respect to derive unbiased estimators from finite-length HMC to continuous un-normalized probability dis- chains. tributions. MCMC estimators typically have Empirically, we observe that two HMC chains targeting higher variance than classical Monte Carlo slightly different stationary distributions still couple with i.i.d. samples due to autocorrelations; approximately when driven by the same randomness. most MCMC research tries to reduce these In this paper, we propose methods that exploit this autocorrelations. In this work, we explore approximate-coupling phenomenon to reduce the vari- a complementary approach to variance re- ance of HMC-based estimators. These methods are duction based on two classical Monte Carlo based on two classical Monte Carlo variance-reduction “swindles”: first, running an auxiliary coupled techniques (sometimes called “swindles”): antithetic chain targeting a tractable approximation to sampling and control variates. the target distribution, and using the aux- iliary samples as control variates; and sec- In the antithetic scheme, we induce anti-correlation ond, generating anti-correlated ("antithetic") between two chains by driving the second chain with samples by running two chains with flipped momentum variables that are negated versions of the randomness. Both ideas have been explored momenta from the first chain.
    [Show full text]
  • Hamiltonian Monte Carlo Within Stan
    Hamiltonian Monte Carlo within Stan Daniel Lee Columbia University, Statistics Department [email protected] BayesComp mc-stan.org 1 Why MCMC? • Have data. • Have a rich statistical model. • No analytic solution. • (Point estimate not adequate.) 2 Review: MCMC • Markov Chain Monte Carlo. The samples form a Markov Chain. • Markov property: Pr(θn+1 j θ1; : : : ; θn) = Pr(θn+1 j θn) • Invariant distribution: π × Pr = π • Detailed balance: sufficient condition: R Pr(θn+1; A) = A q(θn+1; θn)dy π(θn+1)q(θn+1; θn) = π(θn)q(θn; θn+1) 3 Review: RWMH • Want: samples from posterior distribution: Pr(θjx) • Need: some function proportional to joint model. f (x, θ) / Pr(x, θ) • Algorithm: Given f (x, θ), x, N, Pr(θn+1 j θn) For n = 1 to N do Sample θˆ ∼ q(θˆ j θn−1) f (x,θ)ˆ With probability α = min 1; , set θn θˆ, else θn θn−1 f (x,θn−1) 4 Review: Hamiltonian Dynamics • (Implicit: d = dimension) • q = position (d-vector) • p = momentum (d-vector) • U(q) = potential energy • K(p) = kinectic energy • Hamiltonian system: H(q; p) = U(q) + K(p) 5 Review: Hamiltonian Dynamics • for i = 1; : : : ; d dqi = @H dt @pi dpi = − @H dt @qi • kinectic energy usually defined as K(p) = pT M−1p=2 • for i = 1; : : : ; d dqi −1 dt = [M p]i dpi = − @U dt @qi 6 Connection to MCMC • q, position, is the vector of parameters • U(q), potential energy, is (proportional to) the minus the log probability density of the parameters • p, momentum, are augmented variables • K(p), kinectic energy, is calculated • Hamiltonian dynamics used to update q.
    [Show full text]
  • Mixed Hamiltonian Monte Carlo for Mixed Discrete and Continuous Variables
    Mixed Hamiltonian Monte Carlo for Mixed Discrete and Continuous Variables Guangyao Zhou Vicarious AI Union City, CA 94587, USA [email protected] Abstract Hamiltonian Monte Carlo (HMC) has emerged as a powerful Markov Chain Monte Carlo (MCMC) method to sample from complex continuous distributions. However, a fundamental limitation of HMC is that it can not be applied to distributions with mixed discrete and continuous variables. In this paper, we propose mixed HMC (M- HMC) as a general framework to address this limitation. M-HMC is a novel family of MCMC algorithms that evolves the discrete and continuous variables in tandem, allowing more frequent updates of discrete variables while maintaining HMC’s ability to suppress random-walk behavior. We establish M-HMC’s theoretical properties, and present an efficient implementation with Laplace momentum that introduces minimal overhead compared to existing HMC methods. The superior performances of M-HMC over existing methods are demonstrated with numerical experiments on Gaussian mixture models (GMMs), variable selection in Bayesian logistic regression (BLR), and correlated topic models (CTMs). 1 Introduction Markov chain Monte Carlo (MCMC) is one of the most powerful methods for sampling from probability distributions. The Metropolis-Hastings (MH) algorithm is a commonly used general- purpose MCMC method, yet is inefficient for complex, high-dimensional distributions because of the random walk nature of its movements. Recently, Hamiltonian Monte Carlo (HMC) [13, 22, 2] has emerged as a powerful alternative to MH for complex continuous distributions due to its ability to follow the curvature of target distributions using gradients information and make distant proposals with high acceptance probabilities.
    [Show full text]
  • Markov Chain Monte Carlo Methods for Estimating Systemic Risk Allocations
    risks Article Markov Chain Monte Carlo Methods for Estimating Systemic Risk Allocations Takaaki Koike * and Marius Hofert Department of Statistics and Actuarial Science, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada; [email protected] * Correspondence: [email protected] Received: 24 September 2019; Accepted: 7 January 2020; Published: 15 January 2020 Abstract: In this paper, we propose a novel framework for estimating systemic risk measures and risk allocations based on Markov Chain Monte Carlo (MCMC) methods. We consider a class of allocations whose jth component can be written as some risk measure of the jth conditional marginal loss distribution given the so-called crisis event. By considering a crisis event as an intersection of linear constraints, this class of allocations covers, for example, conditional Value-at-Risk (CoVaR), conditional expected shortfall (CoES), VaR contributions, and range VaR (RVaR) contributions as special cases. For this class of allocations, analytical calculations are rarely available, and numerical computations based on Monte Carlo (MC) methods often provide inefficient estimates due to the rare-event character of the crisis events. We propose an MCMC estimator constructed from a sample path of a Markov chain whose stationary distribution is the conditional distribution given the crisis event. Efficient constructions of Markov chains, such as the Hamiltonian Monte Carlo and Gibbs sampler, are suggested and studied depending on the crisis event and the underlying loss distribution. The efficiency of the MCMC estimators is demonstrated in a series of numerical experiments. Keywords: systemic risk measures; conditional Value-at-Risk (CoVaR); capital allocation; copula models; quantitative risk management 1.
    [Show full text]
  • The Geometric Foundations of Hamiltonian Monte Carlo 3
    Submitted to Statistical Science The Geometric Foundations of Hamiltonian Monte Carlo Michael Betancourt, Simon Byrne, Sam Livingstone, and Mark Girolami Abstract. Although Hamiltonian Monte Carlo has proven an empirical success, the lack of a rigorous theoretical understanding of the algo- rithm has in many ways impeded both principled developments of the method and use of the algorithm in practice. In this paper we develop the formal foundations of the algorithm through the construction of measures on smooth manifolds, and demonstrate how the theory natu- rally identifies efficient implementations and motivates promising gen- eralizations. Key words and phrases: Markov Chain Monte Carlo, Hamiltonian Monte Carlo, Disintegration, Differential Geometry, Smooth Manifold, Fiber Bundle, Riemannian Geometry, Symplectic Geometry. The frontier of Bayesian inference requires algorithms capable of fitting complex mod- els with hundreds, if not thousands of parameters, intricately bound together with non- linear and often hierarchical correlations. Hamiltonian Monte Carlo (Duane et al., 1987; Neal, 2011) has proven tremendously successful at extracting inferences from these mod- els, with applications spanning computer science (Sutherland, P´oczos and Schneider, 2013; Tang, Srivastava and Salakhutdinov, 2013), ecology (Schofield et al., 2014; Terada, Inoue and Nishihara, 2013), epidemiology (Cowling et al., 2012), linguistics (Husain, Vasishth and Srinivasan, 2014), pharmacokinetics (Weber et al., 2014), physics (Jasche et al., 2010; Porter and Carr´e,
    [Show full text]
  • Probabilistic Programming in Python Using Pymc3
    Probabilistic programming in Python using PyMC3 John Salvatier1, Thomas V. Wiecki2 and Christopher Fonnesbeck3 1 AI Impacts, Berkeley, CA, United States 2 Quantopian Inc, Boston, MA, United States 3 Department of Biostatistics, Vanderbilt University, Nashville, TN, United States ABSTRACT Probabilistic programming allows for automatic Bayesian inference on user-defined probabilistic models. Recent advances in Markov chain Monte Carlo (MCMC) sampling allow inference on increasingly complex models. This class of MCMC, known as Hamiltonian Monte Carlo, requires gradient information which is often not readily available. PyMC3 is a new open source probabilistic programming framework written in Python that uses Theano to compute gradients via automatic differentiation as well as compile probabilistic programs on-the-fly to C for increased speed. Contrary to other probabilistic programming languages, PyMC3 allows model specification directly in Python code. The lack of a domain specific language allows for great flexibility and direct interaction with the model. This paper is a tutorial-style introduction to this software package. Subjects Data Mining and Machine Learning, Data Science, Scientific Computing and Simulation Keywords Bayesian statistic, Probabilistic Programming, Python, Markov chain Monte Carlo, Statistical modeling INTRODUCTION Probabilistic programming (PP) allows for flexible specification and fitting of Bayesian Submitted 9 September 2015 statistical models. PyMC3 is a new, open-source PP framework with an intuitive and Accepted 8 March 2016 readable, yet powerful, syntax that is close to the natural syntax statisticians use to Published 6 April 2016 describe models. It features next-generation Markov chain Monte Carlo (MCMC) Corresponding author Thomas V. Wiecki, sampling algorithms such as the No-U-Turn Sampler (NUTS) (Hoffman & Gelman, [email protected] 2014), a self-tuning variant of Hamiltonian Monte Carlo (HMC) (Duane et al., 1987).
    [Show full text]
  • Modified Hamiltonian Monte Carlo for Bayesian Inference
    Modified Hamiltonian Monte Carlo for Bayesian Inference Tijana Radivojević1;2;3, Elena Akhmatskaya1;4 1BCAM - Basque Center for Applied Mathematics 2Biological Systems and Engineering Division, Lawrence Berkeley National Laboratory 3DOE Agile Biofoundry 4IKERBASQUE, Basque Foundation for Science July 26, 2019 Abstract The Hamiltonian Monte Carlo (HMC) method has been recognized as a powerful sampling tool in computational statistics. We show that performance of HMC can be significantly improved by incor- porating importance sampling and an irreversible part of the dynamics into a chain. This is achieved by replacing Hamiltonians in the Metropolis test with modified Hamiltonians, and a complete momen- tum update with a partial momentum refreshment. We call the resulting generalized HMC importance sampler—Mix & Match Hamiltonian Monte Carlo (MMHMC). The method is irreversible by construc- tion and further benefits from (i) the efficient algorithms for computation of modified Hamiltonians; (ii) the implicit momentum update procedure and (iii) the multi-stage splitting integrators specially derived for the methods sampling with modified Hamiltonians. MMHMC has been implemented, tested on the popular statistical models and compared in sampling efficiency with HMC, Riemann Manifold Hamiltonian Monte Carlo, Generalized Hybrid Monte Carlo, Generalized Shadow Hybrid Monte Carlo, Metropolis Adjusted Langevin Algorithm and Random Walk Metropolis-Hastings. To make a fair com- parison, we propose a metric that accounts for correlations among samples and weights,
    [Show full text]
  • The Metropolis–Hastings Algorithm C.P
    The Metropolis{Hastings algorithm C.P. Robert1;2;3 1Universit´eParis-Dauphine, 2University of Warwick, and 3CREST Abstract. This article is a self-contained introduction to the Metropolis- Hastings algorithm, this ubiquitous tool for producing dependent simula- tions from an arbitrary distribution. The document illustrates the principles of the methodology on simple examples with R codes and provides entries to the recent extensions of the method. Key words and phrases: Bayesian inference, Markov chains, MCMC meth- ods, Metropolis{Hastings algorithm, intractable density, Gibbs sampler, Langevin diffusion, Hamiltonian Monte Carlo. 1. INTRODUCTION There are many reasons why computing an integral like Z I(h) = h(x)dπ(x); X where dπ is a probability measure, may prove intractable, from the shape of the domain X to the dimension of X (and x), to the complexity of one of the functions h or π. Standard numerical methods may be hindered by the same reasons. Similar difficulties (may) occur when attempting to find the extrema of π over the domain X . This is why the recourse to Monte Carlo methods may prove unavoidable: exploiting the probabilistic nature of π and its weighting of the domain X is often the most natural and most efficient way to produce approximations to integrals connected with π and to determine the regions of the domain X that are more heavily weighted by π. The Monte Carlo approach (Hammersley and Handscomb, 1964; Rubinstein, 1981) emerged with computers, at the end of WWII, as it relies on the ability of producing a large number of realisations of a random variable distributed according to a given distribution, arXiv:1504.01896v3 [stat.CO] 27 Jan 2016 taking advantage of the stabilisation of the empirical average predicted by the Law of Large Numbers.
    [Show full text]
  • Continuous Relaxations for Discrete Hamiltonian Monte Carlo
    Continuous Relaxations for Discrete Hamiltonian Monte Carlo Yichuan Zhang, Charles Sutton, Amos Storkey Zoubin Ghahramani School of Informatics Department of Engineering University of Edinburgh University of Cambridge United Kingdom United Kingdom [email protected], [email protected] [email protected], [email protected] Abstract Continuous relaxations play an important role in discrete optimization, but have not seen much use in approximate probabilistic inference. Here we show that a general form of the Gaussian Integral Trick makes it possible to transform a wide class of discrete variable undirected models into fully continuous systems. The continuous representation allows the use of gradient-based Hamiltonian Monte Carlo for inference, results in new ways of estimating normalization constants (partition functions), and in general opens up a number of new avenues for in- ference in difficult discrete systems. We demonstrate some of these continuous relaxation inference algorithms on a number of illustrative problems. 1 Introduction Discrete undirected graphical models have seen wide use in natural language processing [11, 24] and computer vision [19]. Although sophisticated inference algorithms exist for these models, including both exact algorithms and variational approximations, it has proven more difficult to develop discrete Markov chain Monte Carlo (MCMC) methods. Despite much work and many recent advances [3], the most commonly used MCMC methods in practice for discrete models are based on Metropolis- Hastings, the effectiveness of which is strongly dependent on the choice of proposal distribution. An appealing idea is to relax the constraint that the random variables of interest take integral values.
    [Show full text]