A Rutgers computer scientist is retracting conference proceedings via an unusual channel: his personal blog.
On April 7, Anand Sarwate wrote that he was retracting a mathematical proof from the proceedings from the 2016 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), after discovering errors that invalidated the result.
He explains in the blog post why the mistake occurred:
In the end, I was rather excited about the empirical performance…and we were close to the deadline so I did not do a proper job checking the technical results in our rush to submit. I’m making this a teachable moment for myself but wanted to make sure there was a public, findable version of this retraction so that people don’t continue to build on our results.
The paper — “Symmetric Matrix Perturbation for Differentially-Private Principal Component Analysis” — was added to IEEE Xplore May 19, 2016. It is an “extended abstract” for a mathematical proof he had been working on with his graduate student, Hafiz Imtiaz. Sarwate wrote in the blog:
The fact that I have taken so long to make this public announcement was due to a misguided effort to adapt the algorithm and run new experiments. Unfortunately, this was not to be.
We reached out to Sarwate to better understand the nature of the problem. Broadly speaking, he and Imtiaz are developing statistical tools for working with private data — meaning, mathematics that won’t betray the exact value of a datum in a dataset by finding the other values.
Sarwate gave us an example about salary data to illustrate the formal definition of privacy:
Suppose I want to guarantee privacy to you while I’m trying survey you and other people. If I publish an average salary, that doesn’t guarantee particular privacy to you, because if everyone else reveals their salary it’d be trivial to infer what you’re making, since you’re the only one missing.
Sarwate and Imtiaz thought they could add random noise into the original data to reach the threshold of privacy.
We wrote a procedure for doing a set of calculations. You can implement that set of calculations and that is fine, but we had an argument that if you do this set of calculations, this algorithm guarantees this much privacy.
However:
The argument saying it guaranteed privacy was not correct.
Sarwate publicly owned up to the mistake because other researchers had been emailing him asking him for his computer code.
Doing nothing technically would have been an option, he told us:
There’s a culture in electrical engineering in which conference publications are not considered real publications. They’re an advertisement for what you’re doing right now.
Most people would would kind of say, ‘It was wrong,’ but it might be like a typo to them and it’ll get addressed when they do a longer, more formal paper.
However, Sarwate thought his incorrect proof was significant enough to warrant something more:
The kind of error it is, is a little bigger than the standard errors are in these conference publications.
On May 10, Sarwate told us he had sent an email to someone within IEEE. The paper is not yet retracted. On May 26, he wrote us again, saying he might create his own erratum:
I got a response from IEEE which says that they issue retractions for violations of the Publishing Principles…I guess what I am likely to do at this point is put something on my website with a version of the paper with the erratum attached.
IEEE tells Retraction Watch that Sarwate first contacted them on May 24 — although that seems to be the day they responded to his May 10th email, not the day he first contacted them — and that they are now following their process for potentially removing content from their site. A few years ago, we reported on the apparent retraction of thousands of conference abstracts from the same site; IEEE was unable to tell us how many, exactly.
Hat tip: Rolf Degen
Like Retraction Watch? Consider making a tax-deductible contribution to support our growth. You can also follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up on our homepage for an email every time there’s a new post, or subscribe to our daily digest. Click here to review our Comments Policy. For a sneak peek at what we’re working on, click here.