Subject: Re: New member intro and questions
From: Lynn Winebarger <>
Date: Sat, 4 Dec 1999 05:07:33 -0500 (EST)

On Sat, 4 Dec 1999, Crispin Cowan wrote:
> Lynn Winebarger wrote:
> "mathematics".  That may be the source of some confusion.  In contrast to
> mathematics, computer science research is very applied.  It is not uncommon for CS
> faculty to take off and start a company that productizes their research.  Some
> famous examples:
   I'm not a particularly big fan of taking open and public research and
then closing it up.  I want a different avenue, one that both allows some
measure of profit and allows the research to benefit everyone.  Perhaps
impossible, but still worth trying to find a solution.

> Basically, I'm asking you to consider that becoming a professor may well be the
> easiest way for you to accomplish your goals.

   Well, it's certainly a well-established path, but I don't know if I'd
call it easy.  It might be the easiest among the possible choices at the
moment, but that doesn't mean it has to remain that way (actually, I'd
allow that it is the easiest way, except it doesn't necessarily encourage
the kind of work I'm interested in, at least in math dept.s).

> On the one hand, we could discuss this more intelligently if we knew what that
> area was.  On the other hand, that may be confidential information that you don't
> wish to disclose publicly right now.

   Oh, it's a general enough area that I don't think talking about it will
eliminate many of the opportunities, besides which the number of people
who can effectively work on the problems isn't all that great.  What I'm
interested in is using the current and future research in statistical
modelling/harmonic analysis (which can be looked at as finding algebraic
structure in infinite data and how to usably look at only finite portions
of it) to bear in problems such as codecs and pattern recognition
(including speech and video) - which is already being done, but finding
new codecs requires a certain amount of mathematical research; developing
methods for cleanly separating different kinds of information from a
single signal (e.g. take an audio signal that has both speech and music,
and separate it into two tracks, each of which can then be more accurately
and effectively compressed using specialized codecs) - I don't believe
the math for this has really been done yet (though you will no doubt find
hacks that attempt to do it); more "risky"/long-term development would be
finding useful ways to incorporate statistical models into programming
languages, _especially_ when control flow structures are capable of being
modelled (i.e. meta-level statistical reasoning about the program) - this
can be done today, but languages and their compilers aren't able to take
advantage of it.
   I can list more applications, if you like.  It comes down to when you
understand the fundamental structure (algebraic/statistical) of the 
information you're interested in, you can start designing things in a much
more intelligent and powerful way.

> >    I was thinking something along the lines of (1) and (2) (or actually 2,
> > since Cygnus sells consulting services), as well as writing
> > books/seminars/etc.  Patrons would definitely be helpful, but I don't know
> > how difficult it would be to convince companies to not be free riders.
> > It's the old prisoner's dilemma problem.
> The primary motive for going this way is greed.
   I don't think this is quite accurate.  It's more about control.
Especially with the current trend with university-corporate combined
research where the sponsors of the research have a strong interest in
retaining IP rights.  And the trend of universities seeing themselves more
as businesses than academies.  If I'm going to "make a deal with the
devil", so to speak, I'm going to try and set the terms to my liking as
much as possible, even if it does mean more work for me.

>  Companies make more money than
> faculty; quite often a lot more.  On the other hand, it is more difficult to do,
> especially with far out ideas.  It's like this:
>    * If you can generate product immediately, it's not research, it's applied
>      development.
>    * If you can generate product in the next year or to, it's marginally
>      research.  It's also quite fundable with venture capital, so you just get
>      some investors and go do it.
>    * If it won't generate product for several years, then it's definitely
>      research.  That being the case, it's hard to get any sane people to invest
>      capital in it.  Rather, you have to seek research funding from research
>      sponsors, i.e. DARPA and the NSF.  This is easier to do from a professor's
>      desk.

    I think the government, either through grants or contracts, as well as
foundations, would be among the most promising ways to seek funding
with the least ethical dilemmas (for me, anyway).  One of the good things
about developing free software is that the division between the latter 2
categories can blur, since secrecy isn't so great a concern.

> >    This is very useful advice (having a deep understanding of how the
> > existing institutes work).
> In particular, you want to be aware of the limiting factors that prevent research
> institutions from doing more than they already do.  Shock number 1:  overhead
> rates.  It turns out that a *cheap* overhead rate is 60%, and conventional is 100%
> to 150%.  I've heard that SRI (the Stanford Research Institute) has an overhead
> rate of over 400%.  I.e. a grant brings in $5, and only $1 of it goes to pay for
> researhc staff and equipment, the rest goes to keeping the beaurocracy in place.
> So you try to propose something modest, i.e. one student for a summer and a few
> percent of a professor's time, and before you know it the budget is $50,000.  $1
> million projects are called "medium" sized.
   Are there any books on this subject?  Keeping in mind that I'm not
talking about one of these huge research centers - rather, I'm thinking
something structured more like a partnership, where the company would be
led by the researchers.

> There will be relatively few people interested in weird new things of any kind.
> Combine two weird new things (your research idea being one, and your business
> model being another) and the cross-product gives you a very small pool of people
> to choose from.
   That's probably as it should be.  I don't think the faint of heart
would do very well at such a firm.