Seasoned metaheuristic practitioners acquire vital knowledge about
algorithms, methodologies and components that are likely to be useful
for solving a particular problem. While rarely explicitly documented,
these higher level sensibilities are often key to the success of a
particular approach.
Traditionally, the decomposition of metaheuristics can be considered
'vertical' in that it describes frameworks (Iterated Local Search,
Evolutionary Computation, Ant Colony Optimization etc). While these
frameworks can be described in "pattern language" terms (see [1]), the
additional motivation behind "Metaheuristic Design Patterns" workshop
is two-fold:
Educational:
in
adopting a 'pattern-based' perspective, we're additional
seeking to decompose good research practice by cutting 'horizontally'
across frameworks and methodologies, looking to abstract out aspects
that have yet to be folded back into the mainstream. This can be seen
as serving an educational purpose: "how best to convey the
practicalities of metaheuristic engineering to the uninitiated?"
Declarative:
the "problem statement" aspect of design pattern format essentially
gives heuristic preconditions for the application of a pattern. This
lends itself well to the process of component selection / generation.
Since this is an essential part of the automated design of
metaheuristics, what we'd really like is to be able to convey this
heuristic information declaratively (i.e. so that even a computer can
understand it). The ultimate goal is to use this information to help
reformulate metaheuristic design as a problem of (potentially dynamic)
component assembly. See [2] for more on this.
Call
for Papers
Design Patterns are structured heuristics which capture the essentials
of recurring practice in an abstract form. Their widespread
use
has revolutionized the software industry [1] and the previous
successful workshop on Metaheuristic Design Patterns (MetaDeeP @ GECCO
2014) has demonstated that similar refactorings are possible
in optimization research.
The structure of a pattern can be succinctly described in the following
format:
Intent:
A
description of the goal behind the pattern and the reason
for using it.
Forces
acting: Pre-conditions for
the application of the pattern.
Structure:
A
description of the abstract mechanism of the pattern.
Consequences:
The
likely results and tradeoffs of pattern application.
Examples:
Concrete use-cases (it is traditional to require at least 3 examples
for an abstraction to be `pattern-worthy').
Scope
We welcome submissions of original and unpublished work that presents
ideas of interest to the optmization community in a `Design
Patterns' format, including (but not limited to) the following
areas:
When
applied programmatically within the
optimization workflow, Design Patterns can be seen as a generalized and
structured form of algorithm selection/hyper-heuristics. Particularly
welcome therefore are submissions that include detail on how their
pattern might be applied in an automated fashion, such as:
To what extent can the
`forces
acting' in a metaheuristic design pattern be expressed programmatically?
Can the `consequences'
of pattern
application be defined in a similar manner?
The
AI community has a long tradition of
expressing rule-based systems and there are many useful mechanisms for
this. Machine learning mechanisms for rule induction are similarly
well-developed, leading to the following open issues in the automated
and adaptive design of metaheuristics:
Which metrics (whether
obtained
online or offline) can be used to express the `Forces acting' for a
pattern?
Which metrics might be
applicable
across a range of patterns? (see [2] for more detail on this).
Submission
instructions
We invite submissions of up to 8 pages in length.
Please see the GECCO 2015 information for authors for further
details. Note that the
review process of the workshop is not
double-blind and
hence authors' information should appear in the paper.
We envisage that while proposals for pattern languages and automated
metaheuristic design may require a fuller discussion of up to 8
pages, suggestions for newly-identified MDPs may be
characterised in
shorter submissions. An example of a submission using the recommended
pattern description format is here.
Papers should be submitted in PDF format to MetaDeeP2015@gmail.com
containing the subject line "Metaheuristic Design Patterns Workshop".
All accepted papers will be presented at the workshop and appear in the
GECCO workshop volume. Proceedings of the workshop will be published in
the ACM Digital Library, and may also be distributed at the conference.
Expanded versions of sufficiently high quality submsisions will be
considered for inclusion in a forthcoming
Springer volume.
Important Dates
Submission deadline: April 8, 2015
Author notification: April 25, 2015
Camera-ready submissions: May 10, 2015
Author registration deadline: May 4,
2015
GECCO-2015: July 11-15, 2015.
Workshop Chairs
Chris Simons - University of the West of England, United Kingdom.
chris.simons -at- uwe.ac.uk
Jerry Swan - University of Stirling, United Kingdom.
jsw –at- cs.stir.ac.uk
Krzysztof Krawiec - Poznan University of Technology, Poland.
krawiec –at- cs.put.poznan.pl
Daniel Tauritz – Missouri University of Science and
Technology,
USA.
tauritzd -at- mst.edu
Jim Smith – University of the West of England, United Kingdom
james.smith -at- uwe.ac.uk
References
[1]
Erich Gamma,
Richard Helm,
Ralph
Johnson, and John Vlissides. Design Patterns: Elements of Reusable
Object-Oriented Software. Addison-Wesley Boston, USA. 1995.