@INPROCEEDINGS{king-et-al:wsc2002,
  TYPE = {peer-reviewed-conference},
  AUTHOR = {Gary W. King and Brent Heeringa and Joe Catalano \
            and David L. Westbrook and Paul Cohen},
  TITLE = {Models of Defeat},
  BOOKTITLE = {Proceedings of the 2002 Winter Simulation Conference},
  PAGES = {928--934},
  YEAR = {2002},
  NOTES = {An earlier version appears in the Proceedings of \
           the Second International Conference on Knowledge \
           Systems for Coalition Operations. Pages 85-90},
  ABSTRACT = {Coalition operations are increasingly effects-based, 
which means they apply force only as necessary to achieve political
and psychological effects. Capture the Flag is a war gaming environment
that includes intelligent, autonomous adversaries.  Our planner
previously focused on two goals: occupying objectives and
attrition. But attrition is a means to the defeat of one's enemies,
not an end in itself. For Capture the Flag to plan for defeat, it
needs a model of defeat. We model the ``capacity for conflict'' as a
leaky bucket: when a unit's bucket is full, it has no more capacity
for conflict and it capitulates. Flow into and out of the bucket is
modulated by several factors including attrition and heroism. The
model is inherently dynamical, so it exhibits the time-dependent behaviors one
observes in real conflicts; for example, identical
attacks will have different effects on capitulation as a function of their
timing.}
}

@INPROCEEDINGS{catalano-et-al:pkdd2006,
  TYPE = {peer-reviewed-conference},
  AUTHOR = {Joe Catalano and Tom Armstrong and Tim Oates},
  TITLE = {Discovering Patterns in Real-valued Time Series},
  BOOKTITLE = {Proceedings of the Tenth European Conference on Principles and Practice of \
   Knowledge Discovery in Databases (PKDD)},
  YEAR = {2006},
  NOTES = {To Appear},
  ABSTRACT = {This paper describes an algorithm for discovering variable
length patterns in real-valued time series. In contrast to most existing
pattern discovery algorithms, ours does not first discretize the data, runs
in linear time, and requires constant memory. These properties are obtained
by sampling the data stream rather than processing all of the
data. Empirical results show that the algorithm performs well on both
synthetic and real data when compared to an exhaustive algorithm.}
}