Local Search for Satisfiability: Algorithms and Applications

aroocarmineAI and Robotics

Oct 29, 2013 (3 years and 7 months ago)


Shaowei Cai
30th August 9.30am
N53 0.62
Local Search for Satisfiability: Algorithms and Applications

Abstract: This report presents a proposal for my PhD studies. First, we motivate the problem of
Satisfiability and its importance in computer science and artificial intelligence. We then describe a
critical review of relevant literature in this field, including state-of-the-art techniques. We identify
some challenges in efficiently solving SAT by local search and its application to real world, including
local search for crafted SAT instances and application SAT instances, as well as local search for partial
MaxSAT instances. We describe the research plan and methodologies to address these issues. We
also present our preliminary works we contributed in this direction.

Zahra Shahabi Kargar
30th August 11am
N53 0.62
: Intelligent Elective Surgery Scheduling
Abstract: This study presents a novel approach for solving real world scheduling problems in
dynamic and uncertain environments such as hospital operating rooms (OR). The uncertainty
associated with different activities, along with conflicting priorities and preferences of the
stakeholders, adds additional layers of complexity to the OR scheduling problem, making this one of
the most challenging real world scheduling domains. Although, robust scheduling and stochastic
scheduling have been shown as two powerful approach in dealing with uncertainty in real world
scheduling problems, both of these approaches are highly dependent on the historic data which
determines the probability distribution of stochastic variable and predictive information about the
future events. Rapid advancements in statistical machine learning techniques provide powerful tools
for precise prediction of future events, offering a new stream for tackling uncertainty in dynamic
environment through the integration of machine learning and optimization techniques. This idea has
been applied with limited successful for addressing uncertainty, but there is significant scope for
improvement in current state of the art in this domain.
In this study we propose an intelligent two stage methodology that integrates predicted information
and historical utilization data with optimization techniques to improve current state of the art of real
world dynamic scheduling problems. The proposed framework will employ a novel integration of
machine learning and stochastic programming to address the underlying uncertainty. We use
hospitals’ operating room as an instance of dynamic and uncertain environment and test bed for our
proposed approach.