=============================
REPORT FOR JULY-DECEMBER 2005
=============================

TECHNICAL ACCOMPLISHMENTS

() Developed a new optimization algorithm for constructing conference
schedules, and integrated it with the previous algorithm. The new
algorithm generates better schedules, and it is about four times
faster than the old algorithm.

() Completed work on the initial version of the advanced Information
Elicitor, which is based on the integration of elicitation with
optimization of the conference schedule. The new algorithm evaluates
the utility of potential elicitation questions, and uses best-first
search to identify questions with the highest utility.

() Tested the Space-Time Module on large-scale conference scenarios,
and compared its performance with that of experienced users, in
collaboration with Aaron Steinfeld. The results confirm that the
system performs better than humans, and that the elicitation helps to
improve the schedule quality.

() Developed a mechanism for generating "common-sense" assumptions
about user preferences and available resources, which is used in the
absence of explicit information.

() Developed theoretical foundations of an algorithm for "contingency
scheduling," which allows construction of alternative schedules for
multiple hypothetical scenarios.


IMPROVEMENTS TO PROTOTYPE

() Integrated the Space-Time Module with the Radar Console, in
collaboration with Othar Hanson and Jordan Hayes, which allows the use
of the Space-Time Module as part of the integrated Radar architecture.

() Developed a library of arithmetic operations on probability
distributions, which improves the accuracy of statistical computations
in the optimization and elicitation algorithms.

() Implemented a basic procedure for checking the syntactic and
semantic correctness of conference-planning scenarios. We are now
working on extensions to this error-checking procedure.


OBJECTIVES FOR NEXT PERIOD

() Extend the optimization algorithm to handle more features of
conference scheduling, including room costs, movable equipment,
services, and vendors; also, continue work on improvements to the
search efficiency.

() Develop a new representation of conference-scheduling scenarios,
which will include an API for direct access from other parts of the
Radar architecture.

() Continue work on the advanced Information Elicitor; in particular,
improve the efficiency of search for the highest-utility questions,
and integrate it with fast heuristic evaluation of questions. Also,
begin work on a mechanism for learning of elicitation strategies.

() Implement an initial version of the "contingency scheduling"
system, which will allow construction of alternative schedules for
multiple world models.

() Develop a mechanism for learning "common-sense" assumptions based
on the available information, and integrate it with the current
mechanism for hand-coding these assumptions.


CONCERNS

() The current representation of conference-scheduling scenarios does
not include an API for access from other parts of Radar, which has
slowed down its integration with the Radar Console, and we are now
working on the development of this API.

() We need to develop GUI tools for editing conference-planning
scenarios, which may require collaboration with GUI researchers.
Currently, the construction of such scenarios requires manual editing
of the underlying XML representation, which is a slow process.