Is D the Answer to the One vs. Two Language High ,Performance Computing Dilemma?

Joseph Rushton Wakeling joseph.wakeling at webdrake.net
Mon Aug 12 04:45:02 PDT 2013


On 08/12/2013 05:57 AM, Andrei Alexandrescu wrote:
> On 8/11/13 4:45 PM, Joseph Rushton Wakeling wrote:
>> On Sunday, 11 August 2013 at 23:37:28 UTC, Andrei Alexandrescu wrote:
>>> That's an odd thing to say seeing as a lot of CS academic research is
>>> ten years ahead of the industry.
>>
>> I would personally venture to say that the publication practises of
>> academia in general and CS in particular have many destructive and
>> damaging aspects, and that industry-academia gap might be narrowed quite
>> a bit if these were addressed.
> 
> Could be improved, sure. Destructive and damaging - I'd be curious for some
> substantiation.

In the case of CS in particular, the publication system is different from much
of academia because it's so strongly based around conferences and conference
proceedings.  I'd say that's damaging in several ways.

First, it means people write to the submission deadline rather than to their
work having reached a satisfactory point of readiness.  All other activities
grind to a halt in the run-up to major conference deadlines -- you see students
and postdocs in particular pulling all-nighters in order to make sure that
everything gets done in time.

Besides the health implications of that, such a last-minute rush has plenty of
scope for making mistakes or introducing errors, errors that will be in the
permanent academic record with little scope for correction (conference
proceedings generally don't carry errata).  There are also more direct sources
of bias -- e.g. if the work is based on user surveys, the chances are all the
people in the lab _not_ working towards a paper deadline will be shanghaied into
completing those surveys, disrupting their own work and also ensuring that the
results are based on a very skewed selection of the population.

This pressure to deliver on deadline something that will be accepted by the
conference can also lead to quite a superficial approach to the existing
literature, with references skimmed quickly in order to find any random phrase
that may support the current piece of work (even though on closer reading it may
actually indicate the opposite).

The second source of damage comes via the conference review process.  Because
conferences are all-or-nothing affairs -- you get accepted or you don't --
there's a strong tendency to submit multiple papers presenting different facets
of essentially the same work to multiple different conferences, just to ensure
that _something_ gets accepted.  That means overwork both for the authors (who
have to write all those extra papers) and also for conference referees, who have
to deal with the resulting excess of papers.

Reviewers are also working to deadlines, and with a lot of papers to assess in a
short space of time (which is very disruptive to their other work), that can
lead to snap and very superficial judgements.  If there's a discrepancy in the
amount of work that has to be done -- e.g. a "yes" means just a "yes", but a
"no" means having to write a detailed report explaining why -- that can lead to
accepting papers simply to lessen the workload.

There are also financial aspects -- because most conferences (understandably)
won't accept papers unless at least one author comes to present, it means that
authors' ability to publish their work can be constrained by their labs' ability
to fund travel, accommodation and conference fees rather than by the quality of
what they've done.

And finally, when all is done and dusted, the proceedings of conferences are
almost invariably then locked up behind a publisher paywall, despite the fact
that almost all the document preparation work is done by authors and conference
volunteers.  How many tech businesses can afford the annual subscriptions to
digital libraries?  (I'm thinking small startups here.)

I suppose you could say that many of these issues are personal/professional
failings of individual researchers or labs, but in my experience these failings
are driven by the pressure to publish conference papers, and young researchers
are pretty much trained to follow these working practices in order to succeed.

What particularly frustrates me about this particular situation is that the
justification for the current system -- that computer science is too fast-moving
for journal publication to keep up with the latest results -- simply doesn't
hold water in an age of electronic publication.  It's habit and professional
career structures, rather than the interests of research communication, that
maintain the current system.

I could go on, but I think these examples will serve as substantiation. :-)


More information about the Digitalmars-d mailing list