## approximate Bayesian inference

Posted in Books, pictures, Statistics, Travel, University life with tags , , , , , , , , on March 23, 2016 by xi'an

Maybe it is just a coincidence, but both most recent issues of Bayesian Analysis have an article featuring approximate Bayesian inference. One is by Daniel Graham and co-authors on Approximate Bayesian Inference for Doubly Robust Estimation, while the other one is by Chris Drovandi and co-authors from QUT on Exact and Approximate Bayesian Inference for Low Integer-Valued Time Series Models with Intractable Likelihoods. The first paper has little connection with ABC. Even though it (a) uses a lot of three letter acronyms [which does not help with speed reading] and (b) relies on moment based and propensity score models. Instead, it relies on Bayesian bootstrap, which suddenly seems to me to be rather connected with empirical likelihood! Except the weights are estimated via a Dirichlet prior instead of being optimised. The approximation lies in using the bootstrap to derive a posterior predictive. I did not spot any assessment or control of the approximation effect in the paper.

“Note that we are always using the full data so avoiding the need to choose a summary statistic” (p.326)

The second paper connects pMCMC with ABC. Plus pseudo-marginals on the side! And even simplified reversible jump MCMC!!! I am far from certain I got every point of the paper, though, especially the notion of dimension reduction associated with this version of reversible jump MCMC. It may mean that latent variables are integrated out in approximate (marginalised) likelihoods [as explicated in Andrieu and Roberts (2009)].

“The difference with the common ABC approach is that we match on observations one-at-a-time” (p.328)

The model that the authors study is an integer value time-series, like the INAR(p) model. Which integer support allows for a non-zero probability of exact matching between simulated and observed data. One-at-a-time as indicated in the above quote. And integer valued tolerances like ε=1 otherwise. In the case auxiliary variables are necessary, the authors resort to the alive particle filter of Jasra et al. (2013), which main point is to produce an unbiased estimate of the (possibly approximate) likelihood, to be exploited by pseudo-marginal techniques. However, unbiasedness sounds less compelling when moving to approximate methods, as illustrated by the subsequent suggestion to use a more stable estimate of the log-likelihood. In fact, when the tolerance ε is positive, the pMCMC acceptance probability looks quite close to an ABC-MCMC probability when relying on several pseudo-data simulations. Which is unbiased for the “right” approximate target. A fact that may actually holds for all ABC algorithms. One quite interesting aspect of the paper is its reflection about the advantage of pseudo-marginal techniques for RJMCMC algorithms since they allow for trans-dimension moves to be simplified, as they consider marginals on the space of interest. Up to this day, I had not realised Andrieu and Roberts (2009) had a section on this aspect… I am still unclear about the derivation of the posterior probabilities of the models under comparison, unless it is a byproduct of the RJMCMC algorithm. A last point is that, for some of the Markov models used in the paper, the pseudo observations can be produced as a random one-time move away from the current true observation, which makes life much easier for ABC and explain why exact simulations can sometimes be produced. (A side note: the authors mention on p.326 that EP is only applicable when the posterior is from an exponential family, while my understanding is that it uses an exponential family to approximate the true posterior.)

## at CIRM [#2]

Posted in Mountains, pictures, Running, Statistics, Travel, University life with tags , , , , , , , , , , , , , , , , , , on March 2, 2016 by xi'an

Sylvia Richardson gave a great talk yesterday on clustering applied to variable selection, which first raised [in me] a usual worry of the lack of background model for clustering. But the way she used this notion meant there was an infinite Dirichlet process mixture model behind. This is quite novel [at least for me!] in that it addresses the covariates and not the observations themselves. I still wonder at the meaning of the cluster as, if I understood properly, the dependent variable is not involved in the clustering. Check her R package PReMiuM for a practical implementation of the approach. Later, Adeline Samson showed us the results of using pMCM versus particle Gibbs for diffusion processes where (a) pMCMC was behaving much worse than particle Gibbs and (b) EM required very few particles and Metropolis-Hastings steps to achieve convergence, when compared with posterior approximations.

Today Pierre Druilhet explained to the audience of the summer school his measure theoretic approach [I discussed a while ago] to the limit of proper priors via q-vague convergence, with the paradoxical phenomenon that a Be(n⁻¹,n⁻¹) converges to a sum of two Dirac masses when the parameter space is [0,1] but to Haldane’s prior when the space is (0,1)! He also explained why the Jeffreys-Lindley paradox vanishes when considering different measures [with an illustration that came from my Statistica Sinica 1993 paper]. Pierre concluded with the above opposition between two Bayesian paradigms, a [sort of] tale of two sigma [fields]! Not that I necessarily agree with the first paradigm that priors are supposed to have generated the actual parameter. If only because it mechanistically excludes all improper priors…

Darren Wilkinson talked about yeast, which is orders of magnitude more exciting than it sounds, because this is Bayesian big data analysis in action! With significant (and hence impressive) results based on stochastic dynamic models. And massive variable selection techniques. Scala, Haskell, Frege, OCaml were [functional] languages he mentioned that I had never heard of before! And Daniel Rudolf concluded the [intense] second day of this Bayesian week at CIRM with a description of his convergence results for (rather controlled) noisy MCMC algorithms.

## efficient approximate Bayesian inference for models with intractable likelihood

Posted in Books, pictures, Statistics, University life with tags , , , , , , , , , , , , on July 6, 2015 by xi'an

Dalhin, Villani [Mattias, not Cédric] and Schön arXived a paper this week with the above title. The type of intractable likelihood they consider is a non-linear state-space (HMM) model and the SMC-ABC they propose is based on an optimised Laplace approximation. That is, replacing the posterior distribution on the parameter θ with a normal distribution obtained by a Taylor expansion of the log-likelihood. There is no obvious solution for deriving this approximation in the case of intractable likelihood functions and the authors make use of a Bayesian optimisation technique called Gaussian process optimisation (GPO). Meaning that the Laplace approximation is the Laplace approximation of a surrogate log-posterior. GPO is a Bayesian numerical method in the spirit of the probabilistic numerics discussed on the ‘Og a few weeks ago. In the current setting, this means iterating three steps

1. derive an approximation of the log-posterior ξ at the current θ using SMC-ABC
2. construct a surrogate log-posterior by a Gaussian process using the past (ξ,θ)’s
3. determine the next value of θ

In the first step, a standard particle filter cannot be used to approximate the observed log-posterior at θ because the conditional density of observed given latent is intractable. The solution is to use ABC for the HMM model, in the spirit of many papers by Ajay Jasra and co-authors. However, I find the construction of the substitute model allowing for a particle filter very obscure… (A side effect of the heat wave?!) I can spot a noisy ABC feature in equation (7), but am at a loss as to how the reparameterisation by the transform τ is compatible with the observed-given-latent conditional being unavailable: if the pair (x,v) at time t has a closed form expression, so does (x,y), at least on principle, since y is a deterministic transform of (x,v). Another thing I do not catch is why having a particle filter available prevent the use of a pMCMC approximation.

The second step constructs a Gaussian process posterior on the log-likelihood, with Gaussian errors on the ξ’s. The Gaussian process mean is chosen as zero, while the covariance function is a Matérn function. With hyperparameters that are estimated by maximum likelihood estimators (based on the argument that the marginal likelihood is available in closed form). Turning the approach into an empirical Bayes version.

The next design point in the sequence of θ’s is the argument of the maximum of a certain acquisition function, which is chosen here as a sort of maximum regret associated with the posterior predictive associated with the Gaussian process. With possible jittering. At this stage, it reminded me of the Gaussian process approach proposed by Michael Gutmann in his NIPS poster last year.

Overall, the method is just too convoluted for me to assess its worth and efficiency without a practical implementation to… practice upon, for which I do not have time! Hence I would welcome any comment from readers having attempted such implementations. I also wonder at the lack of link with Simon Wood‘s Gaussian approximation that appeared in Nature (2010) and was well-discussed in the Read Paper of Fearnhead and Prangle (2012).

## Combining Particle MCMC with Rao-Blackwellized Monte Carlo Data Association

Posted in Books, Statistics, University life with tags , , , , , , , , , , , , on October 10, 2014 by xi'an

This recently arXived paper by Juho Kokkala and Simo Särkkä mixes a whole lot of interesting topics, from particle MCMC and Rao-Blackwellisation to particle filters, Kalman filters, and even bear population estimation. The starting setup is the state-space hidden process models where particle filters are of use. And where Andrieu, Doucet and Hollenstein (2010) introduced their particle MCMC algorithms. Rao-Blackwellisation steps have been proposed in this setup in the original paper, as well as in the ensuing discussion, like recycling rejected parameters and associated particles. The beginning of the paper is a review of the literature in this area, in particular of the Rao-Blackwellized Monte Carlo Data Association algorithm developed by Särkkä et al. (2007), of which I was not aware previously. (I alas have not followed closely enough the filtering literature in the past years.) Targets evolve independently according to Gaussian dynamics.

In the description of the model (Section 3), I feel there are prerequisites on the model I did not have (and did not check in Särkkä et al., 2007), like the meaning of targets and measurements: it seems the model assumes each measurement corresponds to a given target. More details or an example would have helped. The extension against the existing appears to be the (major) step of including unknown parameters. Due to my lack of expertise in the domain, I have no notion of the existence of similar proposals in the literature, but handling unknown parameters is definitely of direct relevance for the statistical analysis of such problems!

The simulation experiment based on an Ornstein-Uhlenbeck model is somewhat anticlimactic in that the posterior on the mean reversion rate is essentially the prior, conveniently centred at the true value, while the others remain quite wide. It may be that the experiment was too ambitious in selecting 30 simultaneous targets with only a total of 150 observations. Without highly informative priors, my beotian reaction is to doubt the feasibility of the inference. In the case of the Finnish bear study, the huge discrepancy between priors and posteriors, as well as the significant difference between the forestry expert estimations and the model predictions should be discussed, if not addressed, possibly via a simulation using the posteriors as priors. Or maybe using a hierarchical Bayes model to gather a time-wise coherence in the number of bear families. (I wonder if this technique would apply to the type of data gathered by Mohan Delampady on the West Ghats tigers…)

Overall, I am slightly intrigued by the practice of running MCMC chains in parallel and merging the outcomes with no further processing. This assumes a lot in terms of convergence and mixing on all the chains. However, convergence is never directly addressed in the paper.

## controlled thermodynamic integral for Bayesian model comparison [reply]

Posted in Books, pictures, Running, Statistics, University life with tags , , , , , , , , , , , , on April 30, 2014 by xi'an

Chris Oates wrotes the following reply to my Icelandic comments on his paper with Theodore Papamarkou, and Mark Girolami, reply that is detailed enough to deserve a post on its own:

Thank you Christian for your discussion of our work on the Og, and also for your helpful thoughts in the early days of this project! It might be interesting to speculate on some aspects of this procedure:

(i) Quadrature error is present in all estimates of evidence that are based on thermodynamic integration. It remains unknown how to exactly compute the optimal (variance minimising) temperature ladder “on-the-fly”; indeed this may be impossible, since the optimum is defined via a boundary value problem rather than an initial value problem. Other proposals for approximating this optimum are compatible with control variates (e.g. Grosse et al, NIPS 2013, Friel and Wyse, 2014). In empirical experiments we have found that the second order quadrature rule proposed by Friel and Wyse 2014 leads to substantially reduced bias, regardless of the specific choice of ladder.

(ii) Our experiments considered first and second degree polynomials as ZV control variates. In fact, intuition specifically motivates the use of second degree polynomials: Let us presume a linear expansion of the log-likelihood in θ. Then the implied score function is constant, not depending on θ. The quadratic ZV control variates are, in effect, obtained by multiplying the score function by θ. Thus control variates can be chosen to perfectly correlate with the log-likelihood, leading to zero-variance estimators. Of course, there is an empirical question of whether higher-order polynomials are useful when this Taylor approximation is inappropriate, but they would require the estimation of many more coefficients and in practice may be less stable.

(iii) We require that the control variates are stored along the chain and that their sample covariance is computed after the MCMC has terminated. For the specific examples in the paper such additional computation is a negligible fraction of the total computational, so that we did not provide specific timings. When non-diffegeometric MCMC is used to obtain samples, or when the score is unavailable in closed-form and must be estimated, the computational cost of the procedure would necessarily increase.

For the wide class of statistical models with tractable likelihoods, employed in almost all areas of statistical application, the CTI we propose should provide state-of-the-art estimation performance with negligible increase in computational costs.

## controlled thermodynamic integral for Bayesian model comparison

Posted in Books, pictures, Running, Statistics, University life with tags , , , , , , , , , , on April 24, 2014 by xi'an

Chris Oates, Theodore Papamarkou, and Mark Girolami (all from the University of Warwick) just arXived a paper on a new form of thermodynamic integration for computing marginal likelihoods. (I had actually discussed this paper with the authors on a few occasions when visiting Warwick.) The other name of thermodynamic integration is path sampling (Gelman and Meng, 1998). In the current paper, the path goes from the prior to the posterior by a sequence of intermediary distributions using a power of the likelihood. While the path sampling technique is quite efficient a method, the authors propose to improve it through the recourse to control variates, in order to decrease the variance. The control variate is taken from Mira et al. (2013), namely a one-dimensional temperature-dependent transform of the score function. (Strictly speaking, this is an asymptotic control variate in that the mean is only asymptotically zero.) This control variate is then incorporated within the expectation inside the path sampling integral. Its arbitrary elements are then calibrated against the variance of the path sampling integral. Except for the temperature ladder where the authors use a standard geometric rate, as the approach does not account for Monte Carlo and quadrature errors. (The degree of the polynomials used in the control variates is also arbitrarily set.) Interestingly, the paper mixes a lot of recent advances, from the zero variance notion of Mira et al. (2013) to the manifold Metropolis-adjusted Langevin algorithm of Girolami and Calderhead (2011), uses as a base method pMCMC (Jasra et al., 2007). The examples processed in the paper are regression (where the controlled version truly has a zero variance!) and logistic regression (with the benchmarked Pima Indian dataset), with a counter-example of a PDE interestingly proposed in the discussion section. I quite agree with the authors that the method is difficult to envision in complex enough models. I also did not see mentions therein of the extra time involved in using this control variate idea.

## ¼th i-like workshop in St. Anne’s College, Oxford

Posted in pictures, Statistics, Travel, University life with tags , , , , , , , , , on March 27, 2014 by xi'an

Due to my previous travelling to and from Nottingham for the seminar and back home early enough to avoid the dreary evening trains from Roissy airport (no luck there, even at 8pm, the RER train was not operating efficiently!, and no fast lane is planed prior to 2023…), I did not see many talks at the i-like workshop. About ¼th, roughly… I even missed the poster session (and the most attractive title of Lazy ABC by Dennis Prangle) thanks to another dreary train ride from Derby to Oxford.