alxdavids

"No comments"

Last month I received my 7th academic rejection, but I also got my grubby little mitts on an acceptance to present at IMACC 2017 (held in Oxford in December). Obviously pleased etc and the venue should be pretty nice before xmas. The work I'll be presenting was joint with Martin Albrecht and Enrique Larraia (who has escaped the academic scene and, as is custom, forfeits his right to a hyperlink).

the paper

The work examines an interpretation of the GGH13 multilinear map where all the known ideals have been removed from encodings. That is, remember a GGH encoding looks like [(x + rg)/z] mod q -- where x is the encoded value taken from ZZ_p for a generator g chosen identically for all k source groups. This amounts to an encoding scheme where the plaintext information is encoded in the LSB of the encoding. Our encodings take the form [(x + r/B)/Z] mod q where B is chosen differently for each different source group. B is also subject to some size restrictions to make zero-testing work out. In summary, the encoding is encoded in the MSB of the encoding.

The aim of this work was to initially try and come up with a fix to the "annihilation attacks" of Miles, Sahai and Zhandry that affect BGK-style obfuscation candidates built from GGH13. Some heuristic fixes have created more candidates built from GGH13 that avoid these attacks, unfortunately we didn't really understand dual-input obfuscation enough to really try and do anything here. So we went straight to the multilinear map, this design stood up for a year of analysis (and probably >100 pages of total write-up).

Unfortunately, about 3 months I dealt a fatal blow to our construction by breaking it with an attack that is a simple generalisation of the original MSZ variant when adapted to the MSB setting. This was a sad day for me, but I'm personally glad that our work never saw the light of day in this form.

If you want you can read the full paper online so I won't detail too many spoilers. But, we altered the theme to an analysis of the insecurities of GGH13 even when all the ideals are taken out.

the reviews

In some ways, the paper is the least interesting aspect of the acceptance that we received for the work. The best bit was that we received the worst academic reviews that I have ever witnessed. By worst, I don't mean intensely critical; I mean literally low quality reviews. Here is one of them:

Review 1

PAPER: 21

TITLE: Notes On GGH13 Without The Presence Of Ideals

Overall evaluation: 1 (weak accept)

----------- Overall evaluation -----------

No comments

I mean this is quite a review. Part of its charm is the reviewers commitment to being wholly non-committal.

Firstly, we have the "weak accept". Don't worry, I'm not going to sit here and pretend that I haven't handed this grade out when I've reviewed other work. IMO, the weak accept is a device for saying

"Hey, I'm not so sure about this paper. They've solved a problem, but it is not a problem I really care about. I've found some things I don't like but I'm going to level with you and just admit that I haven't found any rejection-level errors and I don't know the topic that well. However, do not take this as a ringing endorsement; I'll hand responsibility to the other reviewers and absolve myself from the debate."

So fine, whatever. This particular reviewer is probably busy; multiple reviews maybe. They take a glance through each paper. They've read the introduction of this one and seen "multilinear maps" and thought "ffs". But they also see that the paper hasn't attempted to take us back on a nostalgic and delusional trip back to the heady days of 2013. A time when you could walk along regent's canal on a balmy summer evening, overhearing young parents whispering about the potential cryptographic impact of indistinguishability obfuscation to their wide-eyed children.

They make it through to the preliminaries and they glance up at the clock. It reads 9:20pm. Their partner looks over and points out that Paul Hollywood has just launched a scathing critique on a lemon meringue cake on the tv. Arguably a more nuanced piece of feedback than any that will be written in this house tonight. They skip the preliminaries.

They hit the later sections and realise what they are now reading is an attack on a scheme. The attack is fairly simple but they don't have time to check the details. They do a CTRL-F verification search for the words "polynomial time" and find them. This is good news, and based on this alone they can sleep easy that they can remove themselves from further blame if any issues are spotted. A final check that some future work is included in the conclusion.

weak accept, done.

They close down their laptop, reviews done an hour before the deadline. It's consistent, you cannot argue. But they've forgotten the comments. The laptop is quickly opened again, a sigh from the presence next to them. They've already moved onto the feature-length documentary on Josef Stalin's favourite pieces of music.

Apologies are vocalised. Quickly they type, but they realise they have nothing to say about the paper. In fact, "No comments" accurately portrays their feelings towards it. They haven't done this before, but these are desperate circumstances. They write it down, it feels good. It's a 'review'; they have reviewed the text and have found nothing to comment about. Job done. The night is still young. A bird is momentarily stirred nearby by the sharp crack of the closing lid, echoing into the quiet darkness.

Review 2

So review 1 was completely unhelpful. But they're trailing in the wake of another reviewer. Note the lack of time that reviewer 1 spent on top of the "shortest reviews I have received" leaderboard. It was immediately ended by reviewer 2, for the same piece of work.

PAPER: 21

TITLE: Notes On GGH13 Without The Presence Of Ideals

Overall evaluation: 2 (accept)

----------- Overall evaluation -----------

-

This is no ordinary adversary that you would expect to face in an academic review. This person sits in the chaotic neutral section of the d&d morality compass.

They join program committees for conferences, but they have a maniacal distaste for the peer reviewing system. There have been too many missed opportunities at the hands of ill-justified criticisms. They care so little for the process that they will actively try to sabotage it. They roll a d6 for each review and assign it the corresponding score. I was lucky this time, but many others haven't been.

Usually, they take great pleasure in writing an incomprehensible review with no relation to the text they've read. But today is different, they only have one review to write. They rolled an accept, now the dungeon master exposes their bidding.

There is only silence on this occasion. There are no words to be spoken, no acronyms that can replace the empty void where well-reasoned criticism/feedback should go. They try to leave a review with a silent body, but the HTML form doesn't accept it. They could leave a single letter but this would detract from the message. A hyphen represents 'nothing' as good as any other character, and it is thus that the shortest review I have ever received is created.

the summary

It's safe to say that the third review consisted of four paragraphs, along with the final resolution.

"Overall an ok paper with an assembly of observations on GGH."

Receiving these reviews inspired me to go over all the previous ones that I have received and compare them with all the previous reviews that I have written.

received

These are the stats that I accumulated from the reviews that I received.

  • 3 accepts
  • 7 rejects
  • Total words (mean): 420.03333
  • Total words (median): 280.5
  • Max: 1520
  • Min: 0
  • Reject words (mean): 537.7619
  • Reject words (median): 497
  • Accept words (mean): 145.33333
  • Accept words (median): 163

The number of acceptances and rejections corresponds to the final decision, there are many more individual rejections -- don't you worry.

written

Stats that I have accumulated for reviews that I have written

  • 2 accepts
  • 7 rejects
  • Total (mean): 1001.22222
  • Total (median): 889
  • Max: 1495
  • Min: 699
  • Reject words (mean): 1068.42857
  • Reject words (median): 1080
  • Accept words (mean): 766
  • Accept words (median): 766

Firstly, yeah I need to increase my acceptance ratio. Secondly, I am participating in an unequal relationship with respect to the length of reviews that I write and receive. This is not necessarily an indicator of review quality, but I would likely appreciate an 800 word review over a 0 word review.

Also, it seems common that acceptance reviews contain less words in them. I guess it doesn't take long to express that you haven't found anything wrong with a paper.

Either way, I'm writing a shit load for reviews and not getting a reciprocal response. I should probably use my time more wisely.

This blog post is 1522 words long. It took like an hour to compile these figures. I could probably start here.