Archive for the Wines Category
Before leaving Helsinki, we arXived [from the Air France lounge!] the paper Jean-Michel presented on Monday at ABCruise in Helsinki. This paper summarises the experiments Louis conducted over the past months to assess the great performances of a random forest regression approach to ABC parameter inference. Thus validating in this experimental sense the use of this new approach to conducting ABC for Bayesian inference by random forests. (And not ABC model choice as in the Bioinformatics paper with Pierre Pudlo and others.)
I think the major incentives in exploiting the (still mysterious) tool of random forests [against more traditional ABC approaches like Fearnhead and Prangle (2012) on summary selection] are that (i) forests do not require a preliminary selection of the summary statistics, since an arbitrary number of summaries can be used as input for the random forest, even when including a large number of useless white noise variables; (b) there is no longer a tolerance level involved in the process, since the many trees in the random forest define a natural if rudimentary distance that corresponds to being or not being in the same leaf as the observed vector of summary statistics η(y); (c) the size of the reference table simulated from the prior (predictive) distribution does not need to be as large as for in usual ABC settings and hence this approach leads to significant gains in computing time since the production of the reference table usually is the costly part! To the point that deriving a different forest for each univariate transform of interest is truly a minor drag in the overall computing cost of the approach.
An intriguing point we uncovered through Louis’ experiments is that an unusual version of the variance estimator is preferable to the standard estimator: we indeed exposed better estimation performances when using a weighted version of the out-of-bag residuals (which are computed as the differences between the simulated value of the parameter transforms and their expectation obtained by removing the random trees involving this simulated value). Another intriguing feature [to me] is that the regression weights as proposed by Meinshausen (2006) are obtained as an average of the inverse of the number of terms in the leaf of interest. When estimating the posterior expectation of a transform h(θ) given the observed η(y), this summary statistic η(y) ends up in a given leaf for each tree in the forest and all that matters for computing the weight is the number of points from the reference table ending up in this very leaf. I do find this difficult to explain when confronting the case when many simulated points are in the leaf against the case when a single simulated point makes the leaf. This single point ends up being much more influential that all the points in the other situation… While being an outlier of sorts against the prior simulation. But now that I think more about it (after an expensive Lapin Kulta beer in the Helsinki airport while waiting for a change of tire on our airplane!), it somewhat makes sense that rare simulations that agree with the data should be weighted much more than values that stem from the prior simulations and hence do not translate much of an information brought by the observation. (If this sounds murky, blame the beer.) What I found great about this new approach is that it produces a non-parametric evaluation of the cdf of the quantity of interest h(θ) at no calibration cost or hardly any. (An R package is in the making, to be added to the existing R functions of abcrf we developed for the ABC model choice paper.)
The second and third days of AISTATS 2016 passed like a blur, with not even the opportunity to write my impressions in real time! Maybe long tapa breaks are mostly to blame for this… In any case, we had two further exciting plenary talks about privacy-preserving data analysis by Kamalika Chaudhuri and crowdsourcing and machine learning by Adam Tauman Kalai. The talk by Kamalika was covering recent results by Kamalika and coauthors about optimal privacy preservation in classification and a generalisation to correlated data, with the neat notion of a Markov Quilt. Other talks that same day also dwelt on this privacy issue, but I could not be . The talk by Adam was full of fun illustrations on humans training learning systems (with the unsolved difficulty of those humans deliberately mis-training the system, as exhibited recently by the short-lived Microsoft Tay experiment).
Both poster sessions were equally exciting, with the addition of MLSS student posters on the final day. Among many, I particularly enjoyed Iain Murray’s pseudo-marginal slice sampling, David Duvenaud’s fairly intriguing use of early stopping for non-parametric inference, Garrett Bernstein’s work on aggregated Markov chains, Ye Wang’s scalable geometric density estimation [with a special bonus for his typo on the University of Turing, instead of Torino], Gemma Moran’s and Chengtao Li’s posters on determinantal processes, and Matej Balog’s Mondrian forests with a Laplace kernel [envisioning potential applications for ABC]. Again, just to mention a few…
The participants [incl. myself] also took one evening off to visit a sherry winery in Jerez, with a well-practiced spiel on the story of the company, with some building designed by Gutave Eiffel, and with a wine-tasting session. As I personally find this type of brandy too strong in alcohol, I am not a big fan of sherry but it was nonetheless an amusing trip! With no visible after-effects the next morning, since the audience was as large as usual for Adam’s talk [although I did not cross a machine-learning soul on my 6am run…]
In short, I enjoyed very much AISTATS 2016 and remain deeply impressed by the efficiency of the selection process and the amount of involvement of the actors of this selection, as mentioned earlier on the ‘Og. Kudos!
Travelling through Seville, I arrived in Càdiz on Sunday night, along with a massive depression [weather-speaking!]. Walking through the city from the station was nonetheless pleasant as this is an town full of small streets and nice houses. If with less churches than Seville! Richard Samworth gave the first plenary talk of AISTATS 2016 with a presentation on random projections for classification. His classifier is based on an average of a large number of linear random projections of the original data where the projections are chosen as minimising the prediction error over a subset of the components. The performances of this approach seem to be consistently higher than for random forests, which makes it definitely worth investigating further. (A related R package is available.)
The following talks that day covered Bayesian optimisation and probabilistic numerics, with Javier Gonzales introducing glasses for Bayesian optimisation in order to solve its myopia (!)—by which he meant predicting the output of the optimisation over n future steps. And a first mention of the Pima Indians by Daniel Hernandez-Lobato in his talk about EP with stochastic gradient steps towards optimisation. (As well as much larger datasets.) And Mark Girolami bringing quasi-Monte Carlo into control variates. A kernel based ABC by Mijung Park, which uses kernels and maximum mean discrepancy to avoid defining summary statistics, and a version of parallel MCMC by Guillaume Basse. Plus another session on deep learning.
As usual with AISTATS conferences, the central activity of the day was the noon poster session, including speakers discussing their paper, and I had several interesting chats about MCMC related topics, with e.g. one alternative notion of ensemble MCMC [centred on estimating the normalising constant].
We awarded the notable student paper awards before the welcoming cocktail: The winners are Bo Dai, Nedelina Teneva, and Ye Wang. And this first day ended up with a companionable evening in a most genuine tapa bar, tasting local blood sausage and local blue cheese. (If you do not mind the corrida theme!)
Last Friday I spent about 24 hours in Roma due to Clara Grazian defending her thesis there, which was awarded the highest PhD degree from both Sapienza Università di Roma and Université Paris-Dauphine. Her thesis was composed of her papers on ABC for integrated likelihood, on Jeffreys priors for mixtures (which sadly was rejected a few weeks ago), and on scoring rules à la Dawid for model choice. Clara was the first student to graduate from the joint graduate program between Sapienza and Paris-Dauphine, and I look forward the graduation of the next students!It was absolutely wonderful to be there, not only to attend the defence with Marilena Barbieri, Fabrizio Leisen, and Brunero Liseo (who was also Clara’s supervisor) and to congratulate Clara on the completion of her thesis, but also to meet [albeit much too briefly] with old friends, to enjoy great Roman food, perfect weather, my usual long run along the Tiber and twelve of its bridges in the glorious Roman morning, and “just” this unique feeling of Roma in Spring…