🚀 go-pugleaf

                    AAAI 1995 Spring Symposium

                   Extending Theories of Action:
              Formal Theory and Practical Applications

              March 27--29, 1995,  Stanford University

[This document can be found via the URL
http://www.cs.ubc.ca/ftp/local/SSS95/type/CFP.html]

The study of action and its representation has played a central role
in AI, and numerous theories of action have been studied within 
different subdisciplines of AI research, as well as other areas. The 
research focus in these different areas --- among them knowledge 
representation, planning, reasoning under uncertainty, control theory 
and decision theory --- has often been motivated by different issues 
and types of problems, or based on different underlying assumptions. 

Theories of action proposed within the KR community typically address 
compact representations and associated problems (such as the frame, 
qualification and ramification problems). The aim is often to provide 
natural and concise methods for specifying system dynamics. Similar 
motivations but very different methods are used by researchers in 
uncertainty, where the use of influence diagrams and belief networks 
are the norm. In contrast, planning systems usually rely on 
representations that can be more easily exploited by planners in a 
computationally manageable fashion. In control and decision theory, 
while actions play a key role, the question of compact representation 
of actions is seldom addressed, the focus directed to reasoning about 
the effects of actions in small state spaces. However, the types of 
actions and domains dealt with in control and decision theory are 
often more general than those discussed in KR and (classical) planning, 
allowing actions with uncertain and partially-observable effects.

The assumptions made and reasoning methods adopted when putting theories
of action into practice are also often quite different. In many control
and decision problems, search through small state spaces is used or
analytical, closed-form solutions are sought, while in planning, search 
methods over the space of partially-ordered plans are frequently used. 
In the uncertainty community the emphasis in both theory and practice is 
on the exploitation of problem structure (as embodied in network
representations) for computational adavantage.

This symposium is intended to bring together researchers in diverse 
fields of AI, including both theoreticians and practitioners. The aim 
is to assess the state of the art, discuss common foundations of action
representation and reasoning, explore future directions in which
current theories of action must be extended, and identify critical
questions for the successful application of action theories.

The technical focus of the proposed symposium will be twofold: evaluating 
the current state of the art, especially identifying the underlying 
issues that drive research in different fields and characterizing the 
common and distinguishing concerns within these fields; and identifying 
relevant research issues for both the advancement of the state of the art 
and the incorporation of these advances into systems and applications. We 
invite contributions addressing all aspects of representing and reasoning 
about action. Specific topics include, but are not limited to:

* Characterizing and evaluating existing approaches.

  What are the key issues addressed by specific theories of action?
  To what extent do current systems successfully solve the central
  problems? What are the strengths and weaknesses of "current" 
  representations? How can models used in one area be used in others? 
  Are there appropriate common models (e.g., state transition
  models)? How do specific representions relate to these models?
  How can compact actions representations be exploited computationally?
  
* Technical issues in extending theories of actions.
  
  What capabilities should theories of action provide?
  What is required to incorporate the following considerations in 
  our theories: Nondeterminism; default/probabilistic effects and 
  qualifications; Exogenous events; Nonprimitive actions (sequences,
  programs); Concurrent actions; Explicit representation of time; 
  Costs and other notions from utility theory.
  
* Putting Theories of Action into Practice
  
  How can the gap between theory and practice be narrowed? What
  compromises are necessary? How and to what extent should
  implementation considerations influence the formulation of theories
  of action?  What are possible near- and long-term applications and 
  their specific needs? What are appropriate measures of success?
     
We invite extended abstracts (max. five pages) of two sorts:
- Technical papers addressing issues or proposing methods for
  representing or reasoning about action (especially contributions 
  that draw from or bring together different approaches). Reports on 
  work in progress are welcome.
- Position papers advocating particular approaches or steps toward
  integrating ideas from different communities

Others interested in attending should submit a statement of interest
(max. two pages) describing his or her research on relevant problems.
Electronic submissions are strongly encouraged and should be sent to
action@cs.ubc.ca. Paper submissions should be sent to

Craig Boutilier 
Department of Computer Science
University of British Columbia
Vancouver, British Columbia
CANADA, V6T 1Z4
cebly@cs.ubc.ca

Inquiries should be directed to either co-chair.
  
Organizing Committee: Craig Boutilier (co-chair), University of 
British Columbia, cebly@cs.ubc.ca; Tom Dean, Brown University;
Mois\'{e}s Goldszmidt (co-chair), Rockwell, moises@rpal.rockwell.com;
Steve Hanks, University of Washington; David Heckerman, Microsoft;
Ray Reiter; University of Toronto

Submissions Due: October 28, 1994
Notification/Invitations: November 30, 1994
Final Material for Working Notes Due: January 20, 1995
Symposuim Dates: March 27--29, 1995