From: <HIDDEN>
Newsgroups: comp.soft-sys.matlab
Subject: Re: Very weird resolution issue. Bug ????? Seriously !!
Date: Mon, 09 Mar 2009 14:23:58 -0500
Organization: NNTP Server
Lines: 33
Message-ID: <gp3qf0$2pf$>
References: <goju5b$nv4$> <> <gonnuu$krq$> <TASrl.72499$uG1.3742@newsfe16.iad> <gopada$bj8$> <Ajisl.2485$eT1.312@newsfe20.iad> <gou61a$goe$>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
NNTP-Posting-Date: Mon, 9 Mar 2009 19:25:21 +0000 (UTC)
X-Notice: Filtered by postfilter v. 0.7.7
Cancel-Lock: sha1:PIIf7wOnacQ7vT/+abLukJEsI50=
User-Agent: Thunderbird (Windows/20080914)
Xref: comp.soft-sys.matlab:523574

Sung Soo Kim wrote:
> ...not sure why this kind of probabilistic correction is necessary
> because even in statistical sense, consistent biased result is (in
> most case) easier to correct than unbiased noisy result ...

The point is that it is an _unbiased_ estimator that is wanted in 
virtually all instances.

That a system may use either a deterministic or probabilistic correction 
to minimize bias is pretty much open to the implementation on all 
computing language standards that I'm aware of--it's what would 
typically be called a "quality of implementation" issue outside the 
scope of the Standard.  Whether it were to be implemented in the 
hardware (more likely firmware) or software is immaterial.

It is, it seems to me, that the most important thing to take away from 
this thread is that such minutiae as to prescribe floating point 
operations to be reproducible to a lsb is simply beyond the range to 
which any current language standard specifies the behavior of 
"conforming" implementations.

Hence, the job falls back on the programmer to not make unwarranted 
assumptions on numerical results.  If one wanted to become truly famous, 
developing an automated numerical analysis tool that could either fix or 
recast ill-conditioned or ill-posed problems into numerical stable ones 
for a wide variety of nontrivial cases would be a surefire way to do so.