androula@ecn.purdue.edu (Ioannis Androulakis) (05/23/91)
My main interest is the implementation of iterative algorithms
in an asynchronous computing environment. There is no doubt
that asynchronicity has many advantages as well as disadvantages.
But, from my point of view, there exists a much more interesting
question that needs to, at least, be explored.
Is it possible to make use of the concept of asynchronicity
and develop a NEW WAY OF THINKING,
a new way of approaching the problems,
that will lead us to completely
new algorithms that will make full use of the power of the
asynchronous mode of operation and that would not be simple
asynchronous implementations of the existing sequential
algorithms?
I feel like the independent operation of many processors
allows several different views of the problem,
from independent angles.
Would it be possible to actually make use of this new
information that does not exist in sequential algorithms?
Any comments on that will be greatly appreciated.
Thank you,
ioannis
androula@ecn.purdue.edu
--
=========================== MODERATOR ==============================
Steve Stevenson {steve,fpst}@hubcap.clemson.edu
Department of Computer Science, comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabellreynolds@park.bu.edu (John Reynolds) (05/23/91)
One way my thinking has changed is that I now consider the use of discriminant functions instead of search trees for classification and decision making. Discriminant functions can tesselate input space in ways that are independent of the sequential ordering of information and they therefore allow greater generalization. Discriminant functions that are simple functions of input variables can be implemented in parallel extremely efficiently.
will@uunet.UU.NET (William Pickles) (05/24/91)
androula@ecn.purdue.edu (Ioannis Androulakis) writes: >Is it possible to make use of the concept of asynchronicity >and develop a NEW WAY OF THINKING, When I first started writing programs in FCP and Strand - I had to remember the local invocation of Murphy's Law -- Any given event will happen at the least appropriate moment. Any set of events will occur in the sequence you least expect. >a new way of approaching the problems, >that will lead us to completely >new algorithms that will make full use of the power of the >asynchronous mode of operation and that would not be simple >asynchronous implementations of the existing sequential >algorithms? I found that the freely parallel semantics of these languages especially Strand ( the eager head matching of FCP leading to a defensive approach to programming) a very liberating process once I accommodated the above Law. >I feel like the independent operation of many processors >allows several different views of the problem, >from independent angles. >Would it be possible to actually make use of this new >information that does not exist in sequential algorithms? I was working mainly in program analysis tools and discovered the ability to write tools which co-operated in the filling in of slots as the program text being analysed worked its way down the pipeline. William Pickles -- =========================== MODERATOR ============================== Steve Stevenson {steve,fpst}@hubcap.clemson.edu Department of Computer Science, comp.parallel Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell
aboulang@BBN.COM (Albert Boulanger) (05/26/91)
In article <1991May23.120018.17449@hubcap.clemson.edu> androula@ecn.purdue.edu (Ioannis Androulakis) writes:
My main interest is the implementation of iterative algorithms
in an asynchronous computing environment. There is no doubt
that asynchronicity has many advantages as well as disadvantages.
But, from my point of view, there exists a much more interesting
question that needs to, at least, be explored.
Is it possible to make use of the concept of asynchronicity
and develop a NEW WAY OF THINKING,
a new way of approaching the problems,
that will lead us to completely
new algorithms that will make full use of the power of the
asynchronous mode of operation and that would not be simple
asynchronous implementations of the existing sequential
algorithms?
I agree. My interest in asynchronous algorithms was sparked by how
much like quantum mechanical systems they are. Conjugate variables in
QM maps to variables being iterated across processors. In fact I did
a orbiting electron simulation with classical dynamics so that the
momentum variables were being updated on one processor and the
position variables on another, asynchronous to each other. This was a
way of making use of the observer/observed interaction effect
well-known to multiprocessing folk and link it to the something real in
the system being simulated. Asynchronous systems are open systems and
the results of computation are partially influenced by the external
coupling environment. This leads to a kind of high-dimensional noise
(you can read this as nondeterminism if you like) available for
"free". This is all very similar to "non-local" hidden variable
theories in QM. Asynchronous behavior of neural nets is a ripe are of
investigation.
Anyway, a good reference to asynchronous iterative algorithms is the book:
"Parallel & Distributed Computation: Numerical Methods"
Dimitri Bertsekas, & John Tsitsiklis
Seeking space-time pictures of an asynchronous universe,
Albert Boulanger
aboulanger@bbn.com