/ 1

STAT 608 II: Monte Carlo methods in Statistics. Winter 2015.

• Web site: ctools.umich.edu.
• Instructor: Yves Atchadé (yvesa@umich.edu); Office: 272 WH. Phone: 734-763-5238.
• Office Hours: Tues. 4-5pm, or by appointment.
• Lectures: Mo-We. 4-5:30pm, 1068 East Hall.
• GSI: Dao Nguyen, nguyenxd@umich.edu. Office Hours: Mo. 10-11am. SLC in Chemistry
Building, 1720.
• Prerequistes: Regression methods; probability and statistical theory at the level of Stat
425/426, Biostat 601/602. Exposure to computer programming strongly recommended.
• Objectives: This course deals with Markov Chain Monte Carlo (MCMC) methods in statis-
tics. The emphasis is on both the theory and the practical implementation of the algorithms.
At the end of the course, the student should be able to understand the building principles
and the practical aspects of most MCMC algorithms routinely used in statistics. The class
will be taught using programs written in Matlab, R or C. Students may write their home-
work solutions using any of these languages. SPSS, SAS and other very high level statistical
packages are not suitable for this course.
• Outline:
Chapter 1 Basic Monte Carlo methods.
Chapter 2 Markov Chains: an introduction.
Chapter 3 Markov Chains Monte Carlo methods: Metropolis-Hastings, Gibbs sampler,
auxiliary variable methods, Adaptive methods.
Chapter 4 Various topics: doubly intractable distributions, Exact-approximate methods,
sequential Monte Carlo, variational approximations.
• Suggested textbooks:
1. Monte Carlo Statistical methods. C. P. Robert and G. Casella, Springer, 2nd Edition,
2004.
2. Inference in hidden Markov models. O. Cappe, E. Moulines, T. Ryden, Springer, 2005.
3. Monte Carlo Methods in Scientific Computing. Jun S. Liu, Springer, 2001.
4. Markov Chains and Stochastic Stability. S. Meyn, R. Tweedie, 2nd Ed. Cambridge
Univ. Press, 2009.
• Grading: Homework (33%), Project (34%), final exam: (33%).
1. The homeworks will be a mix of theoretical questions and computer programming.
Groups of two students are allowed.
2. The final exam will evaluate the understanding of the algorithms studied in the course.

imsart ver. 2005/10/19 file: syllabus.tex date: March 30, 2015

 / 2

3. The goal of the project is to get students to study research papers in computational
statistics that uses algorithms discussed in the course as building blocks. Each student
should choose one such paper. Groups of two students are allowed. A typical simulation
study might implement one of the algorithms proposed in the chosen paper and inves-
tigate some of the conclusions of the paper. A student might also choose to examine an
unexplored special case or extension, or apply the algorithm of the paper to a similar
problem. The final report submitted should not exceed 15 pages and should include an
introduction, a description of the algorithm, a description of the simulation study to
be performed, the results of the simulation study, a discussion, a conclusion, and the
references. The report should also include a brief review of the statistical context.

Potential papers for the project.

1. C. Andrieu et G. Roberts (2009). The pseudo-marginal approach for efficient Monte Carlo
computations. The Annals of Statistics 37 (2). 697–725.
2. M. Hoffman, D. Blei, J. Paisley, and C. Wang (2013). Stochastic variational inference.
Journal of Machine Learning Research, 14:1303-1347.
3. C. Wang and D. Blei (2013). Variational inference in nonconjugate models. Journal of
Machine Learning Research, 14:1005-1031.
4. A. Raftery , X. Niu , P. Hoff & K. Yeung (2012): Fast Inference for the Latent Space
Network Model Using a Case-Control Approximate Likelihood, Journal of Computational
and Graphical Statistics, 21:4, 901-919.
5. S. Hohna, A. Drummond (2012). Guided Tree Topology Proposals for Bayesian Phylogenetic
Inference Syst Biol (2012) 61 (1): 1-11.
6. S. Goel, M. Salganik (2009). Respondent-driven sampling as Markov chain Monte Carlo.
Statist. Med. 2009; 28:2202–2229.
7. J. Vanhatalo, A. Vehtari. Sparse Log-Gaussian Processes via MCMC for Spatial Epidemi-
ology. JMLR: Workshop and Conference Proceedings 1: 73-89.
8. J. Martin, L. Wilcox, C. Burstedde, O. Ghattas (2012). A stochastic Newton MCMC Method
for Large-Scale Statistical Inverse problems with Applications to Seismic inversion. SIAM
J. SCI. COMPUT. Vol. 34, No. 3, pp. A1460–A1487.
9. Antti Solonen, Heikki Haario, and Marko Laine (2014). Randomize-then-Optimize: a method
for sampling from posterior distributions in nonlinear inverse problems. SIAM Journal on
Scientific Computing, Volume 36, Issue 4, pp. A1359-C399.
10. M. Pereyra (2014). Proximal Markov chain Monte Carlo algorithms. arXiv:1306.0187.
11. A. Schreck, G. Fort, S. Le corff, E. Moulines (2013). A shrinkage-thresholding Metropolis
adjusted Langevin algorithm for Bayesian variable selection.

imsart ver. 2005/10/19 file: syllabus.tex date: March 30, 2015