Mark Devaney

AAAI Conference 1998 Conference Paper

Needles in a Haystack: Plan Recognition in Large Spatial Domains Involving Multiple Agents

• Mark Devaney

While plan recognition research has been applied to a wide variety of problems, it has largely madeidentical assumptions about the number of agents participating in the plan, the observability of the plan execution process, and the scale of the domain. Wedescribe a method for plan recognition in a real-world domain involving large numbers of agents performing spatial maneuversin concert under conditions of limited observability. These assumptions differ radically from those traditionally madein plan recognition and produce a problemwhichcombinesaspects of the fields of plan recognition, pattern recognition, and object tracking. Wedescribe our initial solution whichborrows and builds uponresearch from each of these areas, employinga pattern-directed approach to recognize individual movementsand generalizing these to produce inferences of large-scale behavior.

AAAI Conference 1998 Short Paper

Plan Recognition in Complex Spatial Domains

• Mark Devaney

ICML Conference 1997 Conference Paper

Efficient Feature Selection in Conceptual Clustering

• Mark Devaney
• Ashwin Ram 0001

AAAI Conference 1994 Short Paper

Dynamically Adjusting Categories to Accommodate Changing Contexts

• Mark Devaney

Concept formation is the process by which generalizations are formed through observation of instances from the environment. These instances are described along a number of attributes, which are selected according to their relevance to the problem or task for which the concepts will be used. The context of a concept learning problem consists of the goals and tasks of the learner, as well as its background knowledge and domain theories and the external environment in which it operates. Context is essential to inductive concept learning for it determines which attributes to use for a given problem out of the infinitely many available, providing a bias for the learner (Mitchell, 1980). Furthermore, context is not a static entity, but is constantly changing, especially in the types of learning tasks faced by humans (e.g. Seifert 1989, Barsalou 1991). As concept formation systems are employed in tasks more typical of natural domains and "real-world" problems, the ability to respond to changing contexts becomes increasingly important.