on anonymisation

An article in the New York Times covering a recent publication in Nature Communications on the ability to identify 99.98% of Americans from almost any dataset with fifteen covariates. And mentioning the French approach of INSEE, more precisely CASD (a branch of GENES, as ENSAE and CREST to which I am affiliated), where my friend Antoine worked for a few years, and whose approach is to vet researchers who want access to non-anonymised data, by creating local working environments on the CASD machines  so that data does not leave the site. The approach is to provide the researcher with a dedicated interface, which “enables access remotely to a secure infrastructure where confidential data is safe from harm”. It further delivers reproducibility certificates for publications, a point apparently missed by the New York Times which advances the lack of reproducibility as a drawback of the method. It also mentions the possibility of doing cryptographic data analysis, again missing the finer details with a lame objection.

“Our paper shows how the likelihood of a specific individual to have been correctly re-identified can be estimated with high accuracy even when the anonymized dataset is heavily incomplete.”

The Nature paper is actually about the probability for an individual to be uniquely identified from the given dataset, which somewhat different from the NYT headlines. Using a copula for the distribution of the covariates. And assessing the model with a mean square error evaluation when what matters are false positives and false negatives. Note that the model need be trained for each new dataset, which reduces the appeal of the claim, especially when considering that individuals tagged as uniquely identified about 6% are not. The statistic of 99.98% posted in the NYT is actually a count on a specific dataset,  the 5% Public Use Microdata Sample files, and Massachusetts residents, and not a general statistic [which would not make much sense!, as I can easily imagine 15 useless covariates] or prediction from the authors’ model. And a wee bit anticlimactic.

unbiased HMC

Jeremy Heng and Pierre Jacob arXived last week a paper on unbiased Hamiltonian Monte Carlo by coupling, following the earlier paper of Pierre and co-authors on debiasing by coupling a few weeks ago. The coupling within the HMC amounts to running two HMC chains with common random numbers, plus subtleties!

“As with any other MCMC method, HMC estimators are justified in the limit of the number of iterations. Algorithms which rely on such asymptotics face the risk of becoming obsolete if computational power keeps increasing through the number of available processors and not through clock speed.”

The main difficulty here is to have both chains meet (exactly) with large probability, since coupled HMC can only bring these chain close to one another. The trick stands in using both coupled HMC and coupled Hastings-Metropolis kernels, since the coupled MH kernel allows for exact meetings when the chains are already close, after which they remain happily and forever together! The algorithm is implemented by choosing between the kernels at random at each iteration. (Unbiasedness follows by the Glynn-Rhee trick, which is eminently well-suited for coupling!) As pointed out from the start of the paper, the appeal of this unbiased version is that the algorithm can be (embarrassingly) parallelised since all processors in use return estimators that are iid copies of one another, hence easily merged into a better estimator.

positions in North-East America

Today I received emails about openings in both Université de Montréal, Canada, and Harvard University, USA:

• Professor in Statistics, Biostatistics or Data Science at U de M, deadline October 30th, 2017, a requirement being proficiency in the French language;
• Tenure-Track Professorship in Statistics at Harvard University, Department of Statistics, details there.

here’s to you Nicolas and Bart

“If it had not been for these things, I might have live out my life talking at street corners to scorning men. I might have died, unmarked, unknown, a failure. Now we are not a failure. This is our career and our triumph. Never in our full life could we hope to do such work for tolerance, for justice, for man’s understanding of man as we now do by accident. Our words—our lives—our pains—nothing! The taking of our lives—lives of a good shoemaker and a poor fish peddler—all! That last moment belongs to us—that agony is our triumph.” B. Vanzetti

Today is the 90th anniversary of the execution of Nicolas Sacco and Bartholomeo Vanzetti, Italian anarchists executed in Boston by the State of Massachusetts after an unfair trial. Impacted by anti-Italian prejudice and the radical political views of the accused. Here’s to you is the leading song of the soundtrack of the 1971 movie Sacco e Vanzetti, song written by Ennio Morricone and Joan Baez, also a song I remember singing endlessly in summer camps, as a teenager in the 70’s…

messages from Harvard

As in Bristol two months ago, where I joined the statistics reading in the morning, I had the opportunity to discuss the paper on testing via mixtures prior to my talk with a group of Harvard graduate students. Which concentrated on the biasing effect of the Bayes factor against the more complex hypothesis/model. Arguing [if not in those terms!] that Occam’s razor was too sharp. With a neat remark that decomposing the log Bayes factor as

log(p¹(y¹,H))+log(p²(y²|y¹,H))+…

meant that the first marginal was immensely and uniquely impacted by the prior modelling, hence very likely to be very small for a larger model H, which would then take forever to recover from. And asking why there was such a difference with cross-validation

log(p¹(y¹|y⁻¹,H))+log(p²(y²|y⁻²,H))+…

where the leave-one out posterior predictor is indeed more stable. While the later leads to major overfitting in my opinion, I never spotted the former decomposition which does appear as a strong and maybe damning criticism of the Bayes factor in terms of long-term impact of the prior modelling.

Other points made during the talk or before when preparing the talk:

1. additive mixtures are but one encompassing model, geometric mixtures could be fun too, if harder to process (e.g., missing normalising constant). Or Zellner’s mixtures (with again the normalising issue);
2. if the final outcome of the “test” is the posterior on α itself, the impact of the hyper-parameter on α is quite relative since this posterior can be calibrated by simulation against limiting cases (α=0,1);
3. for the same reason the different rate of accumulation near zero and one  when compared with a posterior probability is hardly worrying;
4. what I see as a fundamental difference in processing improper priors for Bayes factors versus mixtures is not perceived as such by everyone;
5. even a common parameter θ on both models does not mean both models are equally weighted a priori, which relates to an earlier remark in Amsterdam about the different Jeffreys priors one can use;
6. the MCMC output also produces a sample of θ’s which behaviour is obviously different from single model outputs. It would be interesting to study further the behaviour of those samples, which are not to be confused with model averaging;
7. the mixture setting has nothing intrinsically Bayesian in that the model can be processed in other ways.