Next: An Architecture to Support Collective Memory
Up: Introduction
Previous: Execution Traces
In this paper we tackle the issue of how a trace from a completed problem-solving
episode is prepared before it is stored in collective memory.
There are at least
three important reasons why trace data must be prepared before it is stored in
collective memory.
- Data may be bad.
-
Independent agents are unpredictable. Time constraints can lead to
inefficiencies in the plan for an activity.
For these reasons, activity traces can
reflect inefficient patterns of agent activity.
- Data may be too detailed.
- Many details
are irrelevant. Reasoning about the data, to
improve and reuse it, can be simplified by removing detail.
- Data may contain multiple ideas.
-
A single trace may be used to generate a large set of hypotheses
about action sequences that represent regularities in cooperation between agents
for a given domain.
Each hypothesis is a segment of a trace that
accomplishes a set of goals.
Over the generations of problem-solving, certain hypotheses, saved in collective
memory, will be more useful than others.
Andrew Garland
Thu Apr 9 13:39:29 EDT 1998