riese@litsuns1.epfl.ch (Marc Riese) (01/20/91)
I think of `successive refinement' as follows:
if a system must solve a problem in a given amount of time, or must be
prepared to give its best solution at any time,
then a successive refinement approach applies various problem solving
techniques, that increase in terms of time requirements and quality of
solution, until the solution is requested, or the allotted calculation
time expires, or all the applicable techniques have been tried.
I have recently seen and heard references to the same idea, but using a
different name ("progressive reasoning", "anytime algorithms",...).
Can anyone tell me:
1. Who came up with the idea? Maybe it's so obvious that it shouldn't be
attributed to anyone.
2. Are there real differences corresponding to the different names given?
No hair-splitting please. Offhand, I would consider any discussion about
the solution algorithms used as hair-splitting.
Marc Riese
Swiss Federal Institute of Technologypluto@mangani.ucsd.edu (Mark Plutowski) (01/22/91)
riese@litsuns1.epfl.ch (Marc Riese) writes: >I think of `successive refinement' as follows: > if a system must solve a problem in a given amount of time, or must be > prepared to give its best solution at any time, > then a successive refinement approach applies various problem solving > techniques, that increase in terms of time requirements and quality of > solution, until the solution is requested, or the allotted calculation > time expires, or all the applicable techniques have been tried. >I have recently seen and heard references to the same idea, but using a >different name ("progressive reasoning", "anytime algorithms",...). >Can anyone tell me: > 1. Who came up with the idea? Maybe it's so obvious that it shouldn't be > attributed to anyone. > 2. Are there real differences corresponding to the different names given? > No hair-splitting please. Offhand, I would consider any discussion about > the solution algorithms used as hair-splitting. >Marc Riese >Swiss Federal Institute of Technology There is a related, but in fundamental ways different, concept in the statistics and econometrics literature, referred to as the "Method of Sieves." This is a form of nonparametric regression, in which a sequence of models are generated, each model becoming more and more accurate with respect to the true underlying function or time series that the model is attempting to approximate or forecast. This sequence of models is generated (sequentially) in response to a sequential stream of data. However, realizations of this theory tend be computationally demanding, and so would be typically (in practice, although not necessarily in theory) infeasible in real-time. If you obtain any answers to your query with regards to how "successive refinement" methods, if you can post or send me a summary of how these techniques work, I can then give you an more detailed run-down of the differences and similarities between the two techniques. -=-= M.E. Plutowski, pluto%cs@ucsd.edu UCSD, Computer Science and Engineering 0114 9500 Gilman Drive La Jolla, California 92093-0114