Saturday, November 28, 2020

Brief items

 Several items of note:

  • Quanta Magazine remains a generally outstanding source of science articles and opinion pieces.  In this opinion column,  high energy theorist Robert Dijkgraaf gives his views on whether we are reaching "the end of physics".  Spoilers:  he thinks not, and condensed matter physics, with its emergence of remarkable phenomena from underlying building blocks, is one reason.
  • Similarly, I should have pointed earlier to this interesting article by Natalie Wolchover, who asked a number of physicists to define what they mean by "a particle".  I understand the mathematical answer ("a particle is an irreproducible representation of the Poincare group", meaning that it's an object defined by having particular numbers describing how it changes or doesn't under translations in time, space, and rotation).  That also lends itself to a nice definition of a quasiparticle (such an object, but one that results from the collective action of underlying degrees of freedom, rather than existing in the universe's vacuum).  As an experimentalist, though, I confess a fondness for other perspectives.
  • Springer Nature has released its approach for handling open access publication.  I don't think I'm alone in thinking that its fee structure is somewhere between absurd and obscene.  It's simply absurd to think that the funding agencies (at least in the US) are going to allow people to budget €9,500 for a publication charge.  That's equivalent to four months of support for a graduate student in my discipline.  Moreover, the publisher is proposing to charge a non-refundable €2,190 fee just to have a manuscript evaluated for "guided open access" at Nature Physics.  Not that I lack confidence in the quality of my group's work, but how could I possibly justify spending that much for a 75% probability of a rejection letter?  Given that they do not pay referees, am I really supposed to believe that finding referees, soliciting reviews, and tracking manuscript progress costs the publisher €2,190 per submission? 
  • It's older news, but this approach to computation is an interesting one.  Cerebras is implementing neural networks in hardware, and they are doing this through wafer-scale processors (!) with trillions (!) of transistors and hundreds of thousands of cores.  There must be some impressive faulty tolerance built into their network training approach, because otherwise I'd be surprised if even the amazing manufacturing reliability of the semiconductor industry would produce a decent yield of these processors.
  • Older still, one of my colleagues brought this article to my attention, about someone trying to come up with a way to play grandmaster-level chess in a short period of time.  I don't buy into the hype, but it was an interesting look at how easy it seems to be now to pick up machine learning coding skills.  (Instead of deeply studying chess, the idea was to find a compact, memorizable/mentally evaluatable decision algorithm for chess based on training a machine learning system against a dedicated chess engine.) 

10 comments:

Anonymous said...

It would be really interesting if someone donated money to establish an endowment for the non-profit AAAS (which publishes Science) to help pay for open access publications. It would really be a game changer if Science could publish open access for $1000-$2000 per article, which would allow them to undercut Nature significantly.

Anonymous said...

This is a backwards solution. The vast majority of publications are paid through taxpayer money. It is the duty of funding agencies to not squander their own budget on publications fees, so they should be directly negotiating a budget with publication companies. This isn't a problem to be solved by some "benevolent" billionaire.

Let's not forget that the taxpayers are the real ones being abused here. Scientists are hurt as well, of course, but let's not forget our responsibility to the public here.

Anonymous said...

Academic publication is such a racket. We write and referee the papers for free, and then pay for their publication and pay again to read them. I can't honestly say that peer review improved anything I've published and would have preferred to just put them on the arxiv and never bother with traditional journals.

Anonymous said...

Science and Nature do employ professional referees and editors. They also have a News and Views section which needs to get paid for somehow. I'm not saying that it should cost $9000 per article but there are costs associated with employing copyeditors, journalists, etc.

Douglas Natelson said...

Anon@9:31, sure, but Springer-Nature is also a for-profit publisher that charges very high fees to institutional subscribers. Yes, there are costs to turning out a high quality product. Still, for "guided open access", I'd really like to see the accounting analysis that says it should cost $2.5K just to get an article in the review pipeline (maybe - it's not even clear that the fee is only charged if it actually goes out to review). Some form of open access is being mandated by governments and agencies; it's not clear to me that this is the way to do it. Preprint servers have many virtues. Professional society publishers have other approaches worth considering.

Anonymous said...

In other news, the protein folding problem was solved...

(not strictly a condensed-matter problem, I know, but still good to mention here)

Pizza Perusing Physicist said...

Anon at 1:50 PM: On the contrary, I would argue that protein folding can very much be viewed as a soft matter and polymer physics problem.

I assume by 'solved' you are referring to DeepMind's recent AI protein structure prediction (https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving-protein-structures)? If so, I definitely agree that the reported results are a huge and major breakthrough (if they hold up after vetting and peer review from experts), but I am not sure I would go so far as to say that the folding problem has been 'solved'. I guess it depends on your definition of a solution. I recommend watching this conversation with William Bialek (start around 1:55 for the part that is relevant to this comment): https://www.youtube.com/watch?v=PRli78QkEfQ .

Anonymous said...

re Springer Nature etc: It's not them, it's us! It seems to me that a big part of the problem is the role we allow these journals to play in building professional success and institutional reputations. I agree wholeheartedly that it's something like outrageous that the taxpayer is paying Nature an additional 10k to publish research with no value added (modulo peer review--but is anyone suggesting it's any better at, say, Nature than anywhere else? I'm really asking--may my next paper improve during peer review!). But would this not be a nonissue if we, the authors, did not think we would advance our careers by publishing in these journals? I am aware that some think that research published in Nature or Science or its ilk is more likely flawed compared to less high-profile journals, whether because mainly anomalous findings are reported or because the authors are under more pressure to find a high-impact result. But how many hiring committees, say, in experimental condensed matter, would look askance at a candidate with two or three "PRL or glossier" publications to their name versus one with only PRBs, no matter how unimpeachable? How many funding reviews?

Douglas Natelson said...

Anon@4:39, you are preaching to the choir. The community has collectively elevated the perception of Science and Nature and their progeny journals. This has many knock-on effects, and very few of them are healthy.

Anonymous said...

I ask out of pure curiousity, but how this all get started (meaning the focus on Nature and Science journals)? In condensed matter at lower I have heard older physicists say that PRL was the "glossy" journal before the year 2000. Somehow that changed, but I would have assumed there was some conscious effort for Nature and Science specifically, as other subfields of physics still don't care much for those two journals.