Knowledge-based Program Supervision for Medical Image
Processing
Framework:
the LAMA platform for KBS development
Back to main research page:
Framework:
the LAMA platform for KBS development
LAMA is a software platform devoted to the generation of KBSs, i.e.
both knowledge base and inference engine design. The platform supplies
a generic environment, composed of common tools that can be shared for
designing different KBSs. LAMA provides designers with a common kernel
of computational building blocks for the composition of different engines
and their corresponding task-oriented knowledge base framework. The LAMA
platform offers a unified framework to compose engines from reusable reasoning
components or reusing (parts of) existing engines. Engines can also
be tested, compared or modified using the platform. The components of the
LAMA component library are finer grainsized than classical problem-solving
methods, thus offering designers the right level of abstraction: independent
from any programming language, while free enough to fine-tune engine behaviour.
As for experts, LAMA provides them with the same graphical interfaces and
knowledge base description language to express their knowledge at the expertise
level.
Architecture of the LAMA platform mainly composed of a library
of reusable components
for engine design and knowledge base development tools.
The platform includes:
BLOCKS, a
library of components for engine design: just above the programming language,
it provides basic structures and instructions, divided into general and
task-specific ones. These instructions correspond to the designer's interface.
We have developed a first version of this library focusing on the complex,
but well-defined PS task.
YAKL, a language
for expert knowledge expression: it provides a ``model-view'' of the expert
knowledge. The language includes general templates for important
concepts, as well as task-oriented ones for PS, such as goals, operators,
arguments, rules, etc.
KB verification
toolkit, associated facilities for knowledge base verification: they
analyse the contents of an expert knowledge base to check its consistency
and completeness, according to the knowledge assumptions (i.e. contents
and properties) enforced by the engine in use.
YAKL parser and
translator: it translates KB contents into programming structures that
the engine can manage.
LIVE, a graphical
user interface: it enables to visualise the contents of a KB and will soon
allow to follow the solving of a processing objective by the engine.
LAd, a communication
manager: it allows engines, knowledge bases, and interfaces to interact
in a uniformed way.
Several PS engines have been built within the LAMA platform:
PEGASE,
an engine developed on the basis of the engine previously realised in our
team for the OCAPI
system. It performs pure hierarchical planning (or skeletal-plan refinement),
and incorporates a failure
handling mechanism, that enables the transmission of problems accross
the hierarchy of operators.
PULSAR,
was developed as a first tentative towards combining hierarchical and dynamic
operator-based planning methods within the same engine, to overcome PEGASE
lack of flexibility.
MedIA,
is presently developed, to meet specific requirements of functional medical
image processing. It integrates three kinds of planning methods in a complementary
way: hierarchical planning, dynamic abstract step solving, and reactive
operator-based plan adaptation.
More
about LAMA:
LAMA is the result of a common effort within the ORION team: Mar
Marcos (Ph.D student), Patrick
Itey (computer science engineer) and myself work on it at present,
John Van
den Elst and Régis
Vincent contributed to it in the past as Ph.D. students. The overall
research is supervised by Sabine Moisan (senior researcher).
LAMA
main page
LAMA papers
Crubézy,
M., Marcos, M., Moisan, S. Experiments in Building Program Supervision
Engines from Reusable Components.
To be published in ECAI'98 Workshop on Applications
of Ontologies and Problem-Solving Methods. (link
to workshop)
Vincent, R., Moisan,
S., Thonnat, M. : A Library for Program Supervision Engines, INRIA
Research Report 3011, october 1996. (Report
in french).
Some related work
Work around generic
tasks, Center for Cognitive Sciences,
Ohio State University, USA.
KREST, based on
the concept of components of expertise, Free University of Brussels, Belgium.
PROTEGE-II, Stanford Medical Informatics
and Knowledge Systems Laboratory,
Stanford University, USA.
SPARK, BURN and FIREFIGHTER, Cambridge
Research Lab., Digital Equipment Corp.,USA.
Par-KAP and other
work, SWI, University of Amsterdam,
The Netherlands and University of Maryland,
Department of Computer Science, USA.
...
Contributions appear in domains such as knowledge
acquisition, knowledge engineering, software engineering and software
reuse.
Monica
Crubézy