Defeasible Workflow, its Computation and Exception handling

1
Defeasible Workflow, its Computation and
Exception handling

• ZongWei Luo, Amit Sheth, John Miller, Krys Kochut
• LSDIS Lab, Univ of Georgia
• Athens, GA 30602

2
Motivation Scenario - Transport of a Very Low
Birth Weight Infant
Georgia
Ambulance
Medical College of Georgia NICU
Monitor the infant
Rural area
Exceptions
Infection
Respiratory
Cardiac Disease
In collaboration with Dr. Karp Dr. Bhatia in
MCG
Suspect
Review data
Initial Assessment
Diagnosis plan
Definitive management plan
Consultation
Initial Data
Consultation
3
Motivation

• Facts about Clinical Decision Making
• Clinical decision making is a process by which
  alternative strategies of care are considered and
  selected.
• Decisions should be clear cut and error-free.
• Human, environmental influences are frequently
  complex, uncertain and difficult to control.
• The nursing and medical professions have
  expressed an interest in the precision and
  objectivity by which decisions are made, while
  simultaneously retaining an individualistic,
  holistic approach to patient management.
• Standardization of management plans is usually
  accompanied by paying less attention to
  differences in patients needs.
• Defeasible workflow is used to facilitate such a
  decision making.

Meier, P et al. Clinical Decision Making in
Neonatal Intensive Care,Grune Stratton, Inc.
4
Defeasible Workflow

• Default Modeling Execution
• common sense rules can be applied by default
• Dynamic Resolution
• conflicting interests and values preference
  relations
• Exception Handling
• emphasis on experience

Defeasible Decision made based on original
knowledge might not be valid in the future when
more information is available Defeasible
workflow here we use this approach to support
adaptive workflow.
5
Default Modeling

• JECA rules
• ECA (Event, Condition and Action)
• Justification
• Provides necessary context, may be null when
  everything is certain
• Example
• Event Risk factors ( ex.heart murmurs)
• Condition related to Cardiac Disease
• Action Initial Assessment
• Justification blood family member of opposite
  sex does not have Cardiac problem

6
Dynamic resolution

• Evaluation Architecture
• Clustered
• Hierarchical
• Dynamic resolution
• Default value
• Preference relationship

Event Risk factors ( ex.heart murmurs) Condition
related to Cardiac Disease Action Initial
Assessment Justification home member of opposite
sex has Cardiac problem Default value home
member of opposite sex does not has Cardiac
problem
7
Exception handling- Design
Exception
initial
T1
T2
execute
System
User
retry
prepare
Ta
failed
done
Response
Timeout
Omission
initial
Exception hierarchy
execute
prepare
commit
abort
8
A quick review of METEOR ORBWork run-time
HOST 7
HOST 8
TASK Manager
TASK
HOST 1
TASK Scheduler
TASK Scheduler
HOST 9
TASK Scheduler
TASK Scheduler
TASK Scheduler
HOST 6
TASK Manager
HOST 2
TASK Manager
TASK
TASK
9
Task Scheduler Managing Multiple Workflow
Instances
Monitor
Recovery System
Transition1
Data Objects
Task
AND
Transition2
Scheduler
Task Manager
Task Manager
Task Manager
Task Manager
...
Task Manager
...
Task Manager
Pending
Done
Task invocation
Task
Task
Running
10
Exception handling- Runtime

• Handlers for well-understood exceptions
• Case-based
• Manual

Competence based hierarchy
Workflow manager Scheduler Task Manger Task
Realization
11
Current Status
under development
12
Conclusion

• Default modeling dynamic resolution - good
  system design principle
• Model evolution and structure changes - Decision
  made during exception handling
• Case-based approach - learning from experience