Shtetl-Optimized

Back in February, I gave a talk with the above title at the Annual MIT Latke-Hamentaschen Debate.  I’m pleased to announce that streaming video of my talk is now available!  (My segment starts about 10 minutes into the video, and lasts for 10 minutes.)  You can also download my PowerPoint slides here.

Out of hundreds of talks I’ve given in my life, on five continents, this is the single talk of which I’m the proudest.

Of course, before you form an opinion about the issue at hand, you should also check out the contributions of my fellow debaters.  On the sadly-mistaken hamentasch side, my favorite presentation was that of mathematician Arthur Mattuck, which starts in at 56 minutes and lasts for a full half hour (!! – the allotted time was only 8 minutes).  Mattuck relates the shapes of latkes and hamentaschen to the famous Kakeya problem in measure theory—though strangely, his final conclusions seem to provide no support whatsoever for the hamentaschen, even on Mattuck’s own terms.

Finally, what if you’re a reader for whom the very words “latke” and “hamentaschen” are just as incomprehensible as the title of this blog?  OK, here are some Cliff Notes:

• Latkes are fried potato pancakes, traditionally eaten by Jews on Hannukah.
• Hamentaschen are triangular fruit-filled cookies, traditionally eaten by Jews on Purim.
• Beginning at the University of Chicago in 1946, many universities around the world have held farcical annual “debates” between faculty members (both Jewish and non-Jewish) about which of those two foods is better.  (The reason I say “farcical” is simply that, as I explain in my talk, the truth has always been overwhelmingly on one side.)  The debaters have invoked everything from feminist theory to particle physics to bolster their case.

Thanks very much to Dean of Admissions Stu Schmill for moderating, and to MIT Hillel for organizing the debate.

Update: Luboš has a new blog post announcing that he finally found a chapter in Quantum Computing Since Democritus that he likes!  Woohoo!  Whether coincidentally or not, the chapter he likes makes exactly the same points about quantum mechanics that I also make in my pro-latke presentation.

On Sunday afternoon, Dana, Lily, and I were in Copley Square in Boston for a brunch with friends, at the Mandarin Oriental hotel on Boylston Street.  As I now recall, I was complaining bitterly about a number of things.  First, I’d lost my passport (it’s since been found).  Second, we hadn’t correctly timed Lily’s feedings, making us extremely late for the brunch, and causing Lily to scream hysterically the entire car ride.  Third, parking (and later, locating) our car at the Prudential Center was a logistical nightmare.  Fourth, I’d recently received by email a profoundly silly paper, claiming that one of my results was wrong based on a trivial misunderstanding.  Fifth … well, there were other things that were bothering me, but I don’t remember what they were.

Then the next day, maybe 50 feet from where we’d been, the bombs went off, three innocent human beings lost their lives and many more were rendered permanently disabled.

Drawing appropriate morals is left as an exercise for the reader.

Update (Friday, 7AM): Maybe the moral is that you shouldn’t philosophize while the suspects are still on the loose. Last night (as you can read anywhere else on the web) an MIT police officer was tragically shot and killed in the line of duty, right outside the Stata Center, by one of the marathon bombers (who turn out to be brothers from Chechnya). After a busy night—which also included robbing a 7-Eleven (visiting a 7-Eleven that was coincidentally also robbed—no novelist could make this stuff up), carjacking a Mercedes two blocks from my apartment, and randomly throwing some more pressure-cooker bombs—one of the brothers was killed; the other one escaped to Watertown. A massive hunt for him is now underway. MIT is completely closed today, as is Harvard and pretty much every other university in the area—and now, it seems, all stores and businesses in the entire Boston area. The streets are mostly deserted except for police vehicles. As for us, we heard the sirens through much of the night, but didn’t know what they were about until this morning. Here’s hoping they catch the second asshole soon.

Another Update (Friday, 9AM): As the sorry details emerge about these Tsarnaev brothers, it occurs to me that there’s another moral we can draw: namely, we can remind ourselves that the Hollywood image of the evil criminal genius is almost entirely a myth. Yes, evil and genius have occasionally been found in the same person (as with a few of the Nazi scientists), but it’s evil and stupidity that are the far more natural allies. Which is the most optimistic statement I can think to make right now about the future of the human race.

Yet More Updates (Friday, 3PM): The whole Boston area is basically a ghost town now, with the streets empty on a beautiful spring day and the sound of helicopters filling the air.  I was just up on my roofdeck to watch, and never saw anything like it.  I can’t help thinking that it sets a terrible precedent to give a couple doofus amateur terrorists the power to shut down an entire metropolitan area.  Meanwhile, Andrew Sullivan points to a spectacularly stupid tweet by one Nate Bell:

I wonder how many Boston liberals spent the night cowering in their homes wishing they had an AR-15 with a hi-capacity magazine?

This sounds like a gun nut projecting his own disturbed psychology onto other people.  I’m not actually scared, but if I was, owning a gun would do nothing whatsoever to make me less scared (quite the contrary).  What would make me think I could win a gunfight against a frothing lunatic—or that I’d want to find out?  When it comes to violence, the only thing that calms my nerves is a democratic state having a near-monopoly on it.

What else?  It was chilling to watch the Tsarnaev brothers’ aunt, the one in Toronto, babble incoherently on TV about how wonderful her nephews were (a striking contrast to the remorseful uncle in Maryland).  If it emerges that anyone else in this family (including the parents, or the older brother’s wife) had any foreknowledge about the killing spree, then I very much hope they’ll face justice as well.

In other news, Lily had an eventful day too: she finally figured out how to squeeze her toy ball with her hands.

A commenter named Blake Stacey pointed me to a talk entitled The Territory Around BQP: Results and Open Problems, which was given at the Perimeter Institute this past Friday, and which I’d had no idea was available on streaming video.  This talk was part of a fantastic workshop called Conceptual Foundations and Foils for Quantum Information Processing, which was about ways of changing the laws of quantum mechanics to get alternative theories that still make some sort of sense, and that might shed new light on the “tried-and-true original.”  In this particular talk, the speaker discusses a large number of ways to make the complexity class BQP (Bounded-Error Quantum Polynomial-Time) “slightly” bigger or smaller.  I’m embarrassed to admit that I watched this particular talk transfixed to the computer screen: I genuinely couldn’t predict how BQP was going to get mutilated next, and I looked forward to finding out.

Years ago, Sasha Razborov taught me one of my all-time favorite jokes.

In the 1960s, a man starts handing out leaflets in Moscow’s Red Square. Needless to say, he’s immediately apprehended by the KGB. On examining the leaflets, however, the KGB agents discover that they’re just blank pieces of paper. “What is the meaning of this?” the agents demand.

“What could I write?” exclaims the man. “It’s so obvious!”

The lesson I’ve learned this week is that the man was wrong. In politics, nothing is ever too obvious.

I promised to blog more about research, and I will.  Unfortunately, in the one week between my world tour and the start of the fall semester, I’ve been spending less time on quantum complexity research than on sleeping on a new mattress that I bought.  This has provided ample time to ponder the following question, which I’ve decided to add to the Shtetl-Optimized Physics for Doofuses series:

Why is a soft bed more comfortable than a hard one?

At first glance, this question seems too doofusy even for a series such as this, which makes its target audience clear.  The trouble is that, while perfectly reasonable-sounding answers immediately suggest themselves, several of those answers can be shown to be wrong.

Let’s start with the most common answer: a soft bed is more comfortable than a hard bed because it molds to your shape.   The inadequacy of this answer can be seen by the following thought experiment: lie on a soft bed, and let it mold to your body.  Then imagine that the bed retains exactly the same molded shape, but is replaced by ceramic.  No longer so comfortable!

Ah, you reply, but that’s because a ceramic bed doesn’t change its shape as you shift positions throughout the night.  But this reply is still inadequate—since even if you’re lying as still as possible, it still seems clear that a soft bed is more comfortable than a hard one.

So it seems any answer needs to start from the observation that, even when you’re lying still, you’re not really lying still: you’re breathing in and out, there are tiny vibrations, etc.  The real point of a soft bed is to create a gentler potential well, which absorbs the shocks that would otherwise be caused by those sorts of small movements.

(I was tempted to say the point is to damp the movements, but that can’t be right: trampolines are designed for minimal damping, yet sleeping on a trampoline could actually be pretty comfortable.  So the essential thing a bed needs to do is simply to make way in response to small movements and vibrations.  How hard the bed tries to spring back to its original shape is a secondary question—the answer to which presumably influences, for example, whether you prefer an innerspring or a memory-foam mattress.)

So then why aren’t beds even softer than they are?  Well, the limit of infinite softness would be a bed that immediately collapsed to nothing when you lay on it, dropping you to the floor.  But even before that limit, a bed that was too soft would give you too much freedom to shift into awkward positions and thereby cause yourself back problems.  This suggests an answer to a question raised by a colleague: is the purpose of a bed to approximate, as well as possible on the earth’s surface, the experience of sleeping in zero gravity?  Unless I’m mistaken, the answer is no.  Sleeping in space would be like sleeping on a bed that was too soft, with the same potential for back problems and so forth.

Given that lying in bed is normally the least active thing we do, I find it ironic that the only reasons we lie in bed in the first place (as opposed to, say, on steel beams) are dynamical: they involve the way the bed responds to continual vibrations and movements.

I’ll be grateful if knowledgeable physicists, physiologists, or sleepers can correct any errors in the above account.  Meantime, the next time your spouse, partner, roommate, parent, etc. accuses you of lounging in bed all afternoon like a comatose dog, you can reply that nothing could be further from the truth: rather, inspired by a post on Shtetl-Optimized, you’re struggling to reconcile your modern understanding of the physics and biology of lying in bed with the prescientific, phenomenal experience of lying in bed, and thereby make yourself into a more enlightened human being.

In a post a year and a half ago, I offered a prize of $200 for proving something called the Generalized Linial-Nisan Conjecture, which basically said that almost k-wise independent distributions fool AC⁰ circuits.  (Go over to that post if you want to know what that means and why I cared about it.)

Well, I’m pleased to report that that’s a particular $200 I’ll never have to pay.  I just uploaded a new preprint to ECCC, entitled A Counterexample to the Generalized Linial-Nisan Conjecture.  (That’s the great thing about research: no matter what happens, you get a paper out of it.)

A couple friends commented that it was wise to name the ill-fated conjecture after other people rather than myself.  (Then again, who the hell names a conjecture after themselves?)

If you don’t feel like downloading the ECCC preprint, but do feel like scrolling down, here’s the abstract (with a few links inserted):

In earlier work, we gave an oracle separating the relational versions of BQP and the polynomial hierarchy, and showed that an oracle separating the decision versions would follow from what we called the Generalized Linial-Nisan (GLN) Conjecture: that “almost k-wise independent” distributions are indistinguishable from the uniform distribution by constant-depth circuits. The original Linial-Nisan Conjecture was recently proved by Braverman; we offered a $200 prize for the generalized version. In this paper, we save ourselves $200 by showing that the GLN Conjecture is false, at least for circuits of depth 3 and higher.
As a byproduct, our counterexample also implies that Π₂^p⊄P^NP relative to a random oracle with probability 1. It has been conjectured since the 1980s that PH is infinite relative to a random oracle, but the best previous result was NP≠coNP relative to a random oracle.
Finally, our counterexample implies that the famous results of Linial, Mansour, and Nisan, on the structure of AC⁰ functions, cannot be improved in several interesting respects.

To dispel any confusion, the $200 prize still stands for the original problem that the GLN Conjecture was meant to solve: namely, giving an oracle relative to which BQP is not in PH.  As I say in the paper, I remain optimistic about the prospects for solving that problem by a different approach, such as an elegant one recently proposed by Bill Fefferman and Chris Umans.  Also, it’s still possible that the GLN Conjecture is true for depth-two AC⁰ circuits (i.e., DNF formulas).  If so, that would imply the existence of an oracle relative to which BQP is not in AM—already a 17-year-old open problem—and net a respectable $100.

The Onion has a new piece—United Airlines Exploring Viability of Stacking Them Like Cordwood—that, as usual, is grossly unrealistic.  If my own experience is any guide, the real United would never waste money on a grated floor for waste disposal, or people to shovel peanuts into a trough.

But The Onion‘s exploration of the geometry of passenger-packing does raise some genuinely interesting questions.  For years, I’ve had this idea to start an airline where, instead of seats, passengers would get personal cubbyholes that were stacked on top of each other like bunk beds.  (I’d make sure the marketing materials didn’t describe them as “coffin-shaped,” though that’s what they would be.)

You could sleep in your cubbyhole—much more easily than in a seat, of course—but you could also read, watch a movie, work on your laptop, or eat (all activities that I don’t mind doing while lying down, and the first two of which I prefer to do lying down).

Besides passenger comfort, my arrangement would have at least two advantages over the standard one:

First, depending on the exact size of the cubbyholes, you could very likely fit more passengers this way, thereby lowering ticket costs.

Second, assuming the cubbyholes were ventilated, you could put little doors on them, thereby giving passengers far more privacy than in a conventional airline.  No more being immiserated by screaming babies or inane conversations, or the B.O. of the person next to you, or reading lights while you’re trying to sleep.  And, as many of you will have noticed, BQP Aarlines could provide amorous couples with a far more comfortable alternative than the bathroom.

So, readers: do you know if any airline has tried something like this?  If not, why not?  Are there strong arguments against it that I haven’t thought of, besides the obvious cultural/psychological ones?  Should I keep my day job?

Here it is.  It’s mostly about the Singularity and the Many-Worlds Interpretation.

(I apologize if Eliezer and I agreed too much, and also apologize for not quite realizing that the sun was going to set while I was speaking.)

And here’s the discussion that already took place over at Eliezer’s blogging-grounds, Less Wrong.

Recently someone asked me how to give funnier talks.  My first response was to recoil at such an insolent question: doesn’t everyone know that at the core of my shtick lies a unique and ineffable je ne sais quoi that can’t be packaged, bottled, or resold?  But the truth was not that I couldn’t give advice; it’s that I didn’t want to.  For if everyone knew how easy it was to keep an audience at least half-awake, how would people like me maintain their edge?  By proving better theorems?  Having something new and relevant and say?  These questions answer themselves.

But because I love you, my readers, so deeply, and because I feel guilty about abandoning you for so long, I shall now publicly deconstruct the main ingredients of seminar humor, insofar as I’ve been able to find them.  (A few ingredients are specific to theoretical computer science, but most are more general.)

1. Make fun of people in the audience.  (Of course, you have to do it in such a way that they’re flattered you’re ripping them and not someone else.)
2. Ridicule bogus claims related to your topic, particularly claims that received wide currency in the popular press.  (To be honest, I do this not so much because it gets laughs—though it does—but as a small service to humanity.  If I can make one budding crackpot think twice before hitting “Submit” on a disproof of Bell’s Theorem, I will not have lived in vain.  Of course, the ridicule should always focus more on ideas than people; and even then, a few in the audience will frown on it, considering it unscientific or unprofessional.  Forty or fifty crackpots ago, I agreed with them.  It’s only experience that hardened me into a vigilante.)
3. Incorporate the audience’s shared experiences into your talk (without making a big deal of it, as if it’s the most natural thing in the world).  For example, when it comes time to trot out an Alice/Bob scenario, have yours wryly comment on a previous talk, an excursion everyone went on, a current event (like an election) that everyone actually cares about more than the talk…
4. Self-deprecate.  (“My first conjecture was falsified.  The following conjecture hasn’t yet been falsified, and is obviously true…”)
5. Say things that recognize and comment on how neurotic the thought-process of theoretical computer scientists really is, by taking that thought-process to extremes.  (“That’s off by a factor of 10^{10^120}, which is only O(1) and is therefore irrelevant.” “For years, people tried unsuccessfully to prove this sort of impossibility result was impossible.  Our result shows the impossibility of their goal.”)
6. If your field is interdisciplinary, the humor potential is almost limitless.  Are you a physicist?  Ridicule the computer scientists.  A computer scientist?  Ridicule the mathematicians.  A mathematician?  Ridicule the economists.  Chances are, enough differences in notation, terminology, assumptions, and underlying goals will arise in the talk to give you a never-ending supply of material.  “Disciplinary humor” is a more refined, intellectual variant of ethnic humor, and is effective for the same reasons.
7. Explain your results in an unusually vivid or graphic way.  (“If, at the moment of your death, your whole life flashed before you in an instant, and if while you were alive you’d performed suitable quantum computations on your own brain, then you could solve Graph Isomorphism in polynomial time.”)  This type of humor is my absolute favorite: on a plot with laughter volume on one axis and scientific content on the other, it’s way out on the upper-right-hand corner.
8. If you’re using PowerPoint, take full advantage of its comic potential: wild animations, text that pops up on the screen to question or even flat-out contradict what you’re saying, a punchline at the bottom of the slide that only gets revealed when you press a key, etc.  I love doing this because I have as much time as I need to “precompute” jokes (though I’ll then often elaborate on them extemporaneously).
9. Banter with the crowd: if someone makes a crack at your expense, always respond, and even escalate the interaction into a “staged fight” (the rest of the audience will love it).  If someone catches you in a mistake, or you don’t know the answer to a question, make a self-deprecating joke that acknowledges the situation even as it wins you sympathy points.
10. Have high energy!  Loud, lots of moving around, emotion in your voice … like you can’t wait to invite everyone along to the most exciting journey in the history of the universe.  Not only is that good practice in general (at the least, it keeps the audience from falling asleep), it also creates a general atmosphere in which it’s okay to laugh at jokes.
11. Pause a few beats before the punchline.  (You can get better at this by watching professional comics.)
12. Experiment!  If a particular joke bombs, drop it from your rotation; if it brings the house down, recycle it in future talks.  Of course, you should drop a joke once it reaches its saturation point, where much of the audience has already heard it in previous talks.  On the other hand, if this particular audience hasn’t yet heard the joke, disregard your own internal sense of its being “tired”: it could go over just as well as the first time, or better.
13. Steal ideas shamelessly from other speakers.  (I mean their humor techniques, not their results.)  Just as importantly, study the lame jokes other speakers use, so as to avoid them.  (For example, I estimate that 94% of quantum computing talks include a heavy-handed comment about someone or something being “in superposition”; this has not yet gotten a laugh.  Or the talks repeat stories about Feynman, Bohr, etc. that everyone in the audience has already heard a thousand times.)
14. Tailor your jokes to the audience’s background.  For instance, I have some jokes that work great in the US, but sink in other countries.  Or work on physicists but not computer scientists, or vice versa.
15. Make jokes about the country you’re visiting.  Of course, this is subject to common sense: I’ve been known to resort to “zed” / “aboot” jokes in Canada, scone / royalty / powdered wig jokes in England, and neutrality / yodeling jokes in Switzerland, but I usually don’t make the first joke that pops into my head when visiting Germany or Austria.
16. Take risks!  Here’s an Umeshism: if some of your jokes don’t flop, then you’re not being bold enough.  Do things that people can’t believe anyone would actually do in a talk.  Most people seem to operate under the assumption that when they’re giving a talk, they have to be less funny than in regular conversation, when the truth is the opposite.  If something comes into your head that’s funny to you, and it passes the most flimsy and cursory of offensiveness checks … out with it, and worry later about the consequences!

Three final remarks.

First, reading over the list, I can’t help but feel sheepish about how much one can do with such a crude and obvious bag of tricks.

Second, I only wish I applied this crude bag more consistently!  Particularly when I have a new result and I’m excited about the proof, I all too often ignore my own advice and lapse into boringness.  But at least I notice I’m doing it, get annoyed at myself, and resolve to be crasser, less mature, and less professional the next time around.

Third, you might feel that adding shtick to your talks makes you “shallow,” that all that should matter is the content of your results.  In the relatively rare case where you’re addressing experts in your own sub-sub-subfield, that’s probably true: you can drop the funny business and get straight to the point.  In all other cases, I’m almost certain the audience will understand your results better if you incorporate some shtick than if you don’t.  But hey—it’s up to you whether you want to address an ideal Platonic audience (“more lemmas! no irrelevant distractions! yes! harder! faster!”) or the actual flesh-and-blood hairless apes who are dozing off in the seminar room while you speak.

A commenter on my last post writes:

Dear Scott, Please keep the focus of your blog.  You have lately been losing science to your blog and started blogging about various loosely related things. One of the ways I subscribed to your blog was because your articles were very computation-oriented. Now you no longer keep the theme. And as you might have heard, shifting topics in your blog will lose your readers.

So today I noticed something bizarre.  A celebrated result in cryptography, due to Goldreich, Goldwasser, and Micali, states that any pseudorandom generator gives rise to a pseudorandom function family.  See Luca’s notes or the original GGM paper for more.

Now I’d always assumed, without thinking about it, that the GGM result “obviously” carries over to the quantum case—so that any pseudorandom generator secure against quantum attack would give rise to a pseudorandom function family secure against quantum attack.  But now that I’m writing a paper that actually relies on this “fact,” I realized I have no idea why it’s true.

Look: in the GGM argument, you start with a pseudorandom generator G:{0,1}ⁿ→{0,1}²ⁿ, and you apply it recursively to produce a family of functions f_s:{0,1}ⁿ→{0,1}ⁿ, where s is the seed.  You then consider a hypothetical polynomial-time algorithm A that distinguished f_s from a truly random function.  You show how you could use A to create a polynomial-time algorithm that distinguished the output of G from a truly random 2n-bit string—thereby contradicting the starting assumption that G was pseudorandom.

The trouble is, the argument relies crucially on the fact that A examines only a polynomial number of outputs of f_s—intuitively so that you can run a hybrid argument, changing the outputs that A actually examines one by one into truly random strings.  But if A is a quantum algorithm, then (duh) it can examine all 2ⁿ outputs of f_s in superposition!  So any argument that depends on “watching A to see which inputs it queries” is toast.

But maybe we can recover the same conclusion in a fancier way?  For at least seven years, I’ve been going around conjecturing the following:

Conjecture (): Let Q be a quantum algorithm that makes T queries to a Boolean input X∈{0,1}^N.  Then for all ε,δ>0, there exists a deterministic classical algorithm that makes poly(T,1/ε,log(1/δ)) queries to X, and that approximates Q’s acceptance probability to within ε on a 1-δ fraction of inputs.

My motivation for Conjecture () had nothing to do with cryptography.  I was interested in whether we could rule out the possibility that P=BQP relative to a random oracle with probability 1.  If Conjecture () holds—and if the classical algorithm is anything like I think it is—then we can’t rule it out, at least not without proving P≠PSPACE or an even stronger separation in the unrelativized world.

It now occurs to me that, if we knew how to prove Conjecture (), then maybe we could push through a quantum GGM argument using similar ideas—that is, by identifying a tiny subset of inputs to f_s that the quantum algorithm’s acceptance probability “really” depends on.  Alas, I have good reason to believe that Conjecture () is hard.

So the task remains: prove a quantum GGM theorem.  Or maybe I’m missing something completely obvious?

PS. The promised report on the QIS conference in Virginia is coming tomorrow.  Take that, future self!

Update (5/3): An anonymous commenter points out that we can use a simpler hybrid argument of Razborov and Rudich—which doesn’t break down in the quantum case—to show that if there exists a PRG that’s secure against 2^{n^Ω(1)}-time quantum adversaries, then there also exists a PRF with polynomial seed length that’s secure against exponential-time quantum adversaries.  That somehow hadn’t occurred to me, and it’s good enough for my purposes.  (Masked cryptographer: emerge ye from the shadows, and claim thy rightful honour in my Acknowledgments!)  On the other hand, the extremely interesting question still stands of whether one can prove a “strong,” GGM-style reduction: from PRGs secure against f(n)-time quantum adversaries to PRFs with linear seed length secure against f(n)^Ω(1)-time quantum adversaries, for any superpolynomial f.