Tutorial on Distributed Constraint Reasoning
AAMAS-2004 version follows. We will soon post the topics for the IJCAI-2005 edition which
will introduce many novelties, specially secure multiparty computations.
Description
Due to the unprecedented expansion of the Internet, a major change occurs
in our life and habits. We can expect that some software agents would work
on our behalf, representing our interests and defending our privacy. A principled
approach to the achievement of this dream is proposed by the Distributed
Constraint Reasoning community which offers a general framework and powerful
competitive techniques to approach these important applications. Distributed
Constraint Satisfaction is the outcome of intensive research performed in
the 80ies and tries to offer equal opportunities to participants in cooperative
or semi-cooperative distributed resource allocation problems. The field
got a clear contour in the beginning of the 90ies when researchers proved
the main theoretical limitations of distributed computations. This was followed
by a decade of research that continuously increases its intensity.
This tutorial will provide an unified view on Distributed Constraint Reasoning,
introducing distributed constraint reasoning systems as semi-cooperative
multi-agent systems and concentrating on the communication and organization
requirements of such systems. The general ideas behind the known distributed
constraint reasoning systems are presented within this multi-agent framework.
For each approach, the requirements, limitations, advantages and disadvantages
of the different categories will be discussed.
Prerequisite knowledge:
The tutorial is targeted at the general AI audience, both academic and
industrial, working in AI fields that employ search at the core of their systems.
It requires only a basic knowledge in standard algorithm schemes, like branch-and-bound
or local optimization techniques.
Table of Content:
- Introduction: Distributed Search and Distributed Constraint
Reasoning
distribution concepts and their motivation, peculiarities and
motivations for distributing or maintaining problems distributed.
- Frameworks for Distributed Constraint Reasoning
frameworks for artificially distributed versus frameworks for
naturally distributed problems. Frameworks for privacy requirements.
- Distributed Consistency Achievement
- Local Consistency
motivations, degrees of consistency, famous techniques for
achieving consistency.
- Efficiency of Distributed Consistency
distributed nature of consistency, sequential nature of the process.
- Algorithms for achieving distributed local consistency
striving for optimality, techniques related to constraint graphs,
techniques for distributing constraints, continuous domains.
- Incomplete Search for DisCSPs
- Why (not) completeness
impossibility results, FA/C distributions, and natural
approaches to resource allocation.
- Efficient Algorithms with(without) structure
structure and synchronisation as slowing factors in distributed
processes, how far are we from completeness.
- Centralized Search and its Distribution
- Centralized backtracking and straightforward operational
distributions
centralized search, look-ahead, reordering, and consistency
maintenance, the synchronization as simple way of distributing
algorithms.
- Improved Distributions
idle processors and parallelism, improved problem distribution
as means of improving the communication.
- Unhindered parallelism in distributed search
real impediments to parallelism, theoretical results for
distribution techniques as solution to lack of parallelism.
- Asynchronous techniques
- Asynchronism
what is asynchronism and why bother, framework and
explaining results, need and drawbacks of ordering.
- Analysis of equivalences to centralized backtracking
phenomenal analysis versus operational analysis, enumeration,
phenomenal extensions of other centralized search techniques.
- Sources of parallelism in asynchronous solving
parallelism and commitment, the roots of committment.
- Properties and Future
- Samples of Applications
open systems for tracking targets, coordinating emitors,
scheduling, auctions.
- Privacy and openness
alternative techniques for maintaining privacy, classification
of approaches to open and dynamic systems.
Last Change: 10/01/2003