content which deviates from the norm [from Pest county]

your GAN is secretly an energy-based model

As I was reading this NeurIPS 2020 paper by Che et al., and trying to make sense of it, I came across a citation to our paper Casella, Robert and Wells (2004) on a generalized accept-reject sampling scheme where the proposal changes at each simulation that sounds surprising if appreciated! But after checking this paper also appears as the first reference on the Wikipedia page for rejection sampling, which makes me wonder if many actually read it. (On the side, we mostly wrote this paper on a drive from Baltimore to Ithaca, after JSM 1999.)

“We provide more evidence that it is beneficial to sample from the energy-based model defined both by the generator and the discriminator instead of from the generator only.”

The paper seems to propose a post-processing of the generator output by a GAN, generating from the mixture of both generator and discriminator, via a (unscented) Langevin algorithm. The core idea is that, if p(.) is the true data generating process, g(.) the estimated generator and d(.) the discriminator, then

p(x) ≈ p⁰(x)∝g(x) exp(d(x))

(The approximation would be exact the discriminator optimal.) The authors work with the latent z’s, in the GAN meaning that generating pseudo-data x from g means taking a deterministic transform of z, x=G(z). When considering the above p⁰, a generation from p⁰ can be seen as accept-reject with acceptance probability proportional to exp[d{G(z)}]. (On the side, Lemma 1 is the standard validation for accept-reject sampling schemes.)

Reading this paper made me realise how much the field had evolved since my previous GAN related read. With directions like Metropolis-Hastings GANs and Wasserstein GANs. (And I noticed a “broader impact” section past the conclusion section about possible misuses with societal consequences, which is a new requirement for NeurIPS publications.)

racism, discrimination and statistics – examining the history [at the RSS]

The Royal Statistical Society is holding an on-line round table on “Racism, discrimination and statistics – examining the history” on 30 October, at 4pm UK time. The chair is RSS President Deborah Ashby and the speakers are

• John Aldrich – chair of the RSS History Section
• Angela Saini – science journalist
• Stephen Senn – Fisher Memorial Trust

voting inequalities in the US

“We’re the only advanced democracy that deliberately discourages people from voting.” Barack Obama

Following a poorly attended local election in France last weekend, over-interpreted by media and political analysts as usual, with poorer categories more likely to abstain, I reflected on the supplementary degree of voting inequality in the US, where active voter suppression and voting discrimination run uncontested by legislative and constitutional bodies. As it happens, even for federal elections, the election laws are state-based, voted by partisan state lawmakers and implemented by equally partisan officials.This means discriminating practices can become part of these laws, including different restrictions on acceptable forms of identification that poorer voters may be unable to purchase, restrictions on voter registration and in particular on active drives for minority registrations, discriminatory closures of voting (poll) places,  as e.g. a single voting place for 600,000 voters, meaning unreachable stations for those without transportation means and those housebound, abusive voter purges by local administrations, e.g., the Interstate Voter Registration Crosscheck System having 99% more chances to remove legitimate than illegitimate voters, lifelong felon disenfranchisement, including for citizens having completed their sentence, some places asking for on-the-spot proof of US citizenship, involving document poorer voters cannot access, mail-in voting discrimination, no worker protection for participating in the vote, which takes place during the week, grossly underfunded poll budgets, leading for instance to hour long polling queues and various mismanagement of the votes, the possibility for National Guard staffing poll stations, and the century long absurdity of gerrymandering, where something like 60 million Americans live in a place where the ruling party has received the minority of the votes in a state election. Not to mention the election by an electoral college of the president where the winner may lag by 3 million votes behind his contender… And running uncontested grossly misleading political adds

 

“the U.S. census needs a different race question”, does it?

“The stated aim — at least for the last half century — [of the census race question] is to help policy makers and demographers assess whether members of different racial groups have equal access to housing, education, employment and other services, as mandated by law.”

A fairly interesting tribune in Science News on the U.S. census race question and the feature that people often self-identify with a category with “doesn’t always match the box someone else might have checked for them”. The discussion focus on failing to protect discriminated groups because people from said discriminated groups do not identify as members of said discriminated groups. Or, because of a genetic ancestry test like 23andme, people from non-discriminated groups do identify as members of a particular discriminated group, e.g., native American Indians. And while there is a separate question on whether or not the respondant is of Hispanic, Latino or Spanish origin, a third of those answering in the affirmative tick the “other race” box in the census. While the sociologist whose work inspired this article calls for different questions in the census, towards a better reflection of actual discrimination, nowhere is the notion of “race” defined or explicited in this paper. Which may be related to the fact that there is no scientifically accepted such definition, as discussed in this UN report. Except all of us belonging to the Homo sapiens sapiens subspecies and descending from common ancestors in Africa.

I thus wonder at the relevance at keeping such a confusing entry in a census: in several European countries including France, it is actually illegal to collect statistics about the race, ethnicity, religion or ancestry. Given the above confusion in the US census and no clear solution to redress the observed biases, discrimination should be fought on sounder bases…