perm filename CONF.2[HPP,DBL] blob
sn#198225 filedate 1976-01-25 generic text, type T, neo UTF8
Tentative description of the HPP miniconference
Most of the time for scheduled talks at the conf. will be broken
down by PROJECT, not by individual. It is the responsibility of each
project to coordinate a meaningful sequence of talks by its members.
Day 1: Each project will present general material, dealing with:
a) The domain of the project; e.g., enough background on the
variables involved in medical diagnosis, and heuristics
used, to enable the listener to understand some details
of why MYCIN was designed the way it is.
b) The motivation for the program; e.g., how humans perform at this
task, what hopes there were in trying to automate it, why
this would be interesting or useful if successful,...
c) The initial design of the program; the organization, the parts of
the task that were emphasized, the parts that were omitted,
the fundamental representations and algorithms developed.
d) Major problems that cropped up which might be of general character;
the potential solutions considered; how they were solved or deferred
New innovations and ideas which might be of use in other projects.
Days 2,3: Each project will present a series of brief talks and discussions,
at a more detailed level, involving the following themes:
a) Design of large systems
separation of programs from data bases
modular representation of knowledge
multiple representations, multiple internal languages, redundancy
b) Management of large systems
debugging large data bases: inconsistency, assigning blame,...
gathering large quantities of expert knowledge, and handling it.
gathering up experts and dealing with them
program interraction with the expert in dealing with the data base
and explaining the activities of the program.
pragmatics for handling huge programs, with huge time demands
c) Heuristic Search in huge, complex spaces
acquiring heuristics (manually) to guide search
if relevant: automatic acquisition; hypothesis formation, theory
formation, rule-learning, induction,...
altering and adding to this heuristic knowledge
assessing the power of the heuristics used
relationship between generating and pruning heuristics
d) A sample session with the system (fake it if not yet running)
show new knowledge being entered in
show the variety of knowledge and abilities of the system
show the system actually solving a problem it was designed for
show the system explaining itself, assigning blame, ...
don't show more bells and whistles than necessary
don't show more internal details than necessary
e) Interaction of the project with hardware
potential uses for distributed computing, parallel processes, etc.
Imagine a factor of 10000 increase in speed and/or memory,
How would this affect the "success" of your present system?
How might you have designed it differently?
Hints to the Lecturers
1) Use common, well-established terms for the concepts you will be
discussing. Don't invent new words for concepts which already have
adequate English words available.
1b) Conversely, if an English word has a common meaning, don't
pervert it to mean something new just during your lecture! If your
"MATCH" function really just tests for equality, then use the word
"EQUALS"!! Don't UNNECESSARILY strain your audience's minds.
2) Try to abstract concepts from your program to a level above what
is merely an implementation detail. One guideline for deciding when
you are being too specific has been suggested by Randy Davis: If your
description uses program variable names for concepts, then it is too
specific.
3) Try to deal with issues of importance; that means don't talk about
details that won't even interest YOU two months from now. Also, try
to rephrase the issues, problems, strategies, etc. into as general
terms as possible.
3b) If you think some central themes have been omitted from those
already given, you're probably right. Tell us all, soon!
4) When you get the questionaires returned to you later this week,
use them to build up a picture of your audience. What do they already
know? What do they really want to hear about? What DON'T they want to
hear? Why are they coming to your talk?
5) Each group should coordinate its members' talks. When each person
has some idea of what he wants to contribute, the group will meet and
sketch a provisional schedule of talks, topics, prerequisites, etc.
This should be done this month! A tentative schedule for the whole
conference will be distributed well in advance. Each talk should
clearly indicate what prerequisite knowledge will be assumed. This
might mean handing out supplementary "dictionaries" beforehand, going
to some general talk the first day, having taken an organic chemistry
course, having worked on that project for two years,... Clearly,
some of these are ridiculous. If you makethe prerequisites explicit,
you will avoid demanding the impossible of your audience.
6) There may be "state of the project" addresses; panel discussions,
seminars,... Think now about what kind of "inter-project" sessions
would produce the best syntheses of ideas.