Category Archives: Anouncement

Three PhD positions in Lille

My collegue and friend Victor Elvira and his collaborator Francois Septier have three vacant PhD positions in Lille.

The first one is a cooperation with industry: the objective is to predict the evolution of the outputs in the supply chain of the industry partner, i.e. quantities in locations of storage or sales. The thesis will be about mathematical modeling of the supply chain and
inference methods for prediction.
The two other positions actually pertain to a recent and an older interest of mine, dynamical systems and importance sampling. You will study importance sampling based methods for probabilistic inference in complex non-linear high-dimensional systems. More specifically, you will work on novel adaptation schemes in order to overcome current limitations of more traditional IS-based techniques in such a challenging context. (Apply fast, or else I might!)

PostDoc at Imperial College London with Sarah Filippi

My wonderful collegue Sarah Filippi has a postdoctoral opening in statistics to work with her at Imperial College. The topic is Bayesian Nonparametric Statistics for Conditional Independence Tests and Causal Inference, and the start date October 2018. Details about the position are available at

I can highly recommend both Sarah as a researcher and this most interesting and important topic.

Talk in Oxford

Today I gave a talk at the Kernel Monday in Oxford, to talk about Kernel Sequential Monte Carlo (slides). Apart from meeting Dino Sejdinovic  and Brooks Paige again, I also had the chance to get Monsieur Doucets input on some questions. The same ones I got input on from Dino and Brooks later on.

Always amazed by how this town is exactly like beloved Tübingen where I did my undergrad – maybe after adding some ingredients that seem very anachronistic (to me). I also keep on thinking (and Oxford obviously triggers that) that Germany has been overdoing it with its comparatively Anti-Elite research policy, a reaction to Nazi focus on elites. I think some more elitism might be necessary for good research, but its a thin red line: elitism clearly breeds randomness. It’s similar to overfitting/high variance: when you can’t actually predict the future merit of people with high accuracy but still act as if you could.

The picture is some pretty college in Oxford. I’m agnostic with respect to its name.

Talks in England

Next week I’ll be touring England, more specifically the Universities of Reading, Warwick, Oxford and UCL. Apart from visiting dear friends, I’ll be giving a poster in Warwick (still not ready!) and talks at all other places.

19.4.2015 2pm Gradient IS and Unadjusted Langevin for IS Univ. of Reading Afternoon on Bayesian Computation, Nike Lecture Theatre, Agriculture Building
22.4.2016 1pm Kernel Sequential Monte Carlo UCL Roberts G08 Sir David Davies LT, Announcement
25.4.2016 3:30pm Kernel Sequential Monte Carlo  Univ. of Oxford Department of Statistics, St Giles building, Seminar Room 1, LG.03

The poster will be on Flyweight evidence estimates at the CRiSM workshop on Estimating Constants on the 20th. Slides and the poster will go up on the Talks page.

Importance Sampling Session at MCQMC 2016

On behalf of the mcqmc 2016 organizing committee I am pleased to accept your proposal.
-Art Owen
I got this nice message from Art yesterday night. My proposal for a session on Advances in Importance Sampling at MCQMC 2016 got accepted. Which is great, as I think the session is made up of strong papers (obviously). This session will almost surely be moderated by Nicolas Chopin.

MCQMC session on Advances in Importance Sampling

The sample size required in Importance Sampling

S. Chatterjee, P. Diaconis

The goal of importance sampling is to estimate the expected value of a given function with respect to a probability measure ν using a random sample of size n drawn from a different probability measure μ. If the two measures μ and ν are nearly singular with respect to each other, which is often the case in practice, the sample size required for accurate estimation is large. In this article it is shown that in a fairly general setting, a sample of size approximately exp(D(ν||μ)) is necessary and sufficient for accurate estimation by importance sampling, where D(ν||μ) is the Kullback–Leibler divergence of μ from ν. In particular, the required sample size exhibits a kind of cut-off in the logarithmic scale. The theory is applied to obtain a fairly general formula for the sample size required in importance sampling for exponential families (Gibbs measures). We also show that the standard variance-based diagnostic for convergence of importance sampling is fundamentally problematic. An alternative diagnostic that provably works in certain situations is suggested.

Generalized Multiple Importance Sampling

V. Elvira, L. Martino, D. Luengo, M. Bugallo

Importance Sampling methods are broadly used to approximate posterior distributions or some of their moments. In its standard approach, samples are drawn from a single proposal distribution and weighted properly. However, since the performance depends on the mismatch between the targeted and the proposal distributions, several proposal densities are often employed for the generation of samples. Under this Multiple Importance Sampling (MIS) scenario, many works have addressed the selection or adaptation of the proposal distributions, interpreting the sampling and the weighting steps in different ways. In this paper, we establish a general framework for sampling and weighing procedures when more than one proposal are available. The most relevant MIS schemes in the literature are encompassed within the new framework, and, moreover novel valid schemes appear naturally. All the MIS schemes are compared and ranked in terms of the variance of the associated estimators. Finally, we provide illustrative examples which reveal that, even with a good choice of the proposal densities, a careful interpretation of the sampling and weighting procedures can make a significant difference in the performance of the method.

Continuous-Time Importance Sampling

K. Łatuszyński, G. Roberts, G. Sermaidis, P. Fearnhead

We will introduce a new framework for sequential Monte Carlo, based on evolving a set of weighted particles in continuous time. This framework can lead to novel versions of existing algorithms, such as Annealed Importance Sampling and the Exact Algorithm for diffusions, and can be used as an alternative to MALA for sampling from a target distribution of interest. These methods are amenable to the use of sub-sampling, which can greatly increase their computational efficiency for big-data applications; and can enable unbiased sampling from a much wider-range of target distributions than existing approaches.