ameen@techunix.BITNET (Ameen Abu Hanna) (11/11/87)
I'm interested in the field of diagnostic ESs that reason from
first principles ('deep', 'model based/driven', 'high_road' ... as they
appear in literature).
My Masters' project is to build an ES shell which, to be a specific ES,
would incorporate a 'deep' model about the diagnosed system .
The modeled system is to be hierarchial, built out of subcomponents ending
up in primitives, it supports the following properties about the model:-
o allows feedback loops ,
o states can be explicitly represented in the primitives' behaviour,
o delay time of a primitive is expressed as a function of some parameters,
o there is a 'library' of components' types.
Examples of modeled systems would be : a digital/analog electronic card,
a car, or other physical systems that meet some requirements.
The main point of the ES is to let it 'learn' by compiling a 'shallower'
expert, this is done by abstracting the behaviour of the components at
different levels: I thought assigning responsibilities to sub_components
(at different abstraction levels), and learning about them would be a good idea.
----
I'm familiar with papers of Davis, Genesereth (I also saw their extended
papers in the book "Qualitative Reasoning about Physical Systems" assembled
by Bobrow, 85') and others appearing in AAAI and IJCAI in recent years,
they rather deal with static systems.
What I'm trying to find are:
o references dealing with ways to attack the diagnostic task when
the diagnosed system is dynamic and similar to the above description.
o references dealing with constructing (automatically) a shallower expert,
inferred from the deep knowledge about the modeled system. (Probably
getting a more 'qualitative' description of behaviour from the more
quantitative description).
o any Ph.D/Masters thesis/project in progress/completed in the above field.
o any one interrested in the above who would like to exchange ideas.
Your help would be highly appreciated !.
If you think you can help, could you please e-mail me at the e-mail address
given below.
if there would be enough responses (a nice surprise, although according to
the rating of the top 40 popular groups, it is estimated that 15,000 readers
exist for the comp.ai group worldwide) I'll summarize to the net,
otherwise I'll e-mail any interrested individual.
I've read (and responded to ) a similar posting by Shyam Chittajullu
(AT&T Bell Labs), but here I look for a more specific field.
Thanks in advance !
ameen abu_hanna
Dept. of Computer Science
Technion,
Haifa Israel.
--------------------------------------------------------------------------------
Domain: ameen@techunix.technion.ac.il UUCP: ...!ucbvax!ameen@techunix.bitnet
BITNET: ameen@techunix ARPANET: ameen%techunix.bitnet@wiscvm.arpa
--------------------------------------------------------------------------------