[net.ai] Seminar - Learning Design Synthesis Expertise

MSIMS@RUTGERS.ARPA (04/19/84)

From:  Michael Sims  <MSIMS@RUTGERS.ARPA>

         [Forwarded from the Rutgers bboard by Laws@SRI-AI.]


Learning  Design  Synthesis  Expertise  by  Harmonizing  Behaviors with
                            Specifications

Speaker:      Masanobu Watanabe <Watanabe@Rutgers.Arpa>
              NEC Corporation, Tokyo, Japan
              Visiting Researcher, Rutgers University

Series:       Machine Learning Brown Bag Seminar
Date:         Wednesday, April 25, 1984, 12:00-1:30
Location:     Hill Center, Room 254


       VEXED is an expert system which supports interactive circuit design.
    VEXED provides suggestions  regarding  alternative  implementations  of
    circuit modules, as well as warnings regarding conflicting constraints.
    The   interactions  between  a  human  designer  and  the  system  give
    opportunities for the system to learn expertise in design synthesis  by
    monitoring  the  human  designer's  response  to  advice offered by the
    system.  From this point of view, there are two interesting cases.  One
    occurs  when  the  designer  ignores the advice of the system.  Another
    occurs when the system cannot provide any advice but the human designer
    can continue his own design.

       The system has to learn as many things as possible  by  analyzing  a
    single  precious  example,  because  it  is difficult for the system to
    obtain many examples from which to form  a  particular  concept.    The
    problem  space in the module decomposition process can be viewed as one
    with both states consisting of a set of modules  and  operators,  which
    will   be  called  implementation  rules.    This  talk  discusses  the
    implementation rule acquisition task which is intended to formulate  an
    implementation rule at an appropriate level of generality by monitoring
    a   designer's   circuit   implementation.    This  task  is  to  learn
    implementation rules (a kind of operator,  but  not  quite  like  LEX's
    operators),  while  LEX's  task  is  to learn heuristics which serve to
    guide useful operators.