Is Self-Adaptation of Selection Pressure and Population Size
Download
Report
Transcript Is Self-Adaptation of Selection Pressure and Population Size
JJ Merelo Guervós
Dept. of Computer
Architecture and Technology
U. of Granada
Six Degrees of Separation to
Darrell Whitley
Session #4
PPSN XI
Reykjavik, Iceland
12/Sept/2006
Everything is connected
2
Improved Squeaky Wheel Optimisation for Driver
Scheduling,
Uwe Aickelin, Edmund K. Burke, Jingpeng Li
SWO is an algorithm based on
a construction-analysisprioritization cycle.
Improved ISWO introduces
selection and mutation within
the solution.
Driver scheduling involves
partitioning
Each component
blocks of
must
work
prove
(1
vehicle
each) into legal shifts
its fitness.
set covering integer linear
programming problem.
3
Search bias in ant colony optimization: on the role of
competition-balanced systems Blum, C.; Dorigo, M.
4
An Evolutionary Approach to the Inference of
Phylogenetic Networks,
Juan Diego Trujillo and Carlos Cotta
Trying to find a phylogenetic network
that models a set of sequences of
molecular data using EAs
Networks include reticulation events
(horizontal transfer, recombination,
hybridization).
Heuristics based genetic operators.
Fitness function based on likelihood.
Network models close to evolutionary
model hidden in the data.
5
Yao, X. Evolving artificial neural networks
(1999) Proceedings of the IEEE, 87 (9), pp. 1423-1447
6
Genetic Algorithm based on Independent Component
Analysis for Global Optimization
Gang Li, Kin Hong Lee, Kwong Sak Leung
ICA projects an n-dimensional set to
a lower-dimensional space.
Since components are independent
of each other, they can be
independently maximized.
Solutions are comparable to other
algorithms, but fewer fitness
evaluations.
7
X. Yao and Y. Liu. Fast evolution strategies
8
When Do Heavy-Tail Distributions Help?
Hansen, Gemperle, Auger, and Koumoutsakos
Cauchy distribution is an example
of heavy-tail.
As opposed to gaussian.
Studies the probability of sampling
a better solution using different
Cauchy distribution.
Anisotropic Cauchy obtains
exceptionally good results on
Rastrigin function.
9
Yao, X., Liu, Y. (1998)
Towards designing artificial neural networks by evolution
10
Neuroevolution with Analog Genetic Encoding
Peter Dürr, Claudio Mattiussi, and Dario Floreano
Recent paper by Banzhaf et al. In Nature ReviewsGenetics request following more closely known
biological facts in EC to convert it into computational
evolution.
This paper uses an encoding for neural nets closer to
real genomes
Based on tokens that represent problem objects.
With operators to match
11
Castillo-Valdivieso, P.A.et al. (2002)
Statistical analysis of the parameters of a neuro-genetic algorithm
12
Assortative mating drastically alters the
magnitude of error thresholds
Gabriela Ochoa and Klaus Jaffe
Beyond the error threshold, evolved
structures cannot be reproduced in
the quasispecies evolution model
(Eigen and Schuster).
In EC, related to the
exploration/exploitation balance.
Assortative mating produces the
highest error threshold, whereas
asexual reproduction produces the
lowest.
13
Eiben, Hinterding, Michalewicz (1999)
Parameter control in evolutionary algorithms
14
Cumulative Step Length Adaptation on Ridge
Functions
Dirk V. Arnold
Ridge functions used to test ES.
This paper studies the
performance of multirecombinative
ES with cumulative step length
adaptation for different ridge
topologies.
15
M. Herdy.
Reproductive isolation as strategy parameter in
hierarchically organized evolution
strategies
16
Self-Adaptation on the Ridge Function Class: First
Results for the Sharp Ridge
Hans-Georg Beyer and Silja Meyer-Nieberg
Different self-adaptation mechanism.
Different ridge: sharp ridge, in this
case.
Different ES: non-recombinative
strategies.
Why could it fail+
17
Arabas, J., Michalewicz, Z., Mulawka, J. (1994)
GAVaPS, A Genetic Algorithm with Varying
Population Size.
18
Self-regulated Population Size in
Evolutionary Algorithms
Carlos Fernandes and Agostinho Rosa
There are many self-regulated
population algorithms: GAVaPS, APGA,
ProFiGa:
Funky acronym not a requisite.
Some based on age.
SRP-EA combines CHC and GAVaPS.
Achieves better success rates, with a
penalty in the number of evaluations.
19
Is Self-Adaptation of Selection Pressure and
Population Size Possible? – a Case Study
A.E. Eiben M.C. Schut A.R. de Wilde
Self-adapting selection operators
and population size can yield as
good or better results than selfadapting operators.
Selection parameters are
encoded in individuals, and a
consensus value is reached.
Self-adapting selection increases
speed.
20
Zitzler, E., Laumanns, M., and Thiele, L. (2001).
SPEA2: Improving the strength
pareto evolutionary algorithm.
21
Solving Multi-Objective Optimisation Problems Using the Potential
Pareto Regions EvolutionaryAlgorithm
Nasreddine Hallam, Graham Kendall and Peter Blanchfield
Introduces Potential Pareto Regions
Evolutionary Algorithm.
The fitness of an individual is equal to
the least improvement needed by that
individual in order to reach a nondominated status.
22
K. Deb, L. Thiele, M. Laumanns, and E. Zitzler (2002).
Scalable multi-objective optimization test
problems.
23
Pareto Set and EMOA Behavior for Simple
Multimodal Multiobjective Functions
Mike Preuss, Boris Naujoks, and Günter Rudolph
Studies the often-disregarded Pareto set.
Changes induced in Pareto set alter the
ability of algorithms to track Pareto Front.
A measure of the quality of solution in the
solution space is needed.
Similar to S-metric in objective space.
24
About Selecting the Personal Best
in Multi-objective Particle Swarm Optimization
Jürgen Branke and Sanaz Mostaghim
Selecting a good guide bodes well for
the future of a particle.
But they can also memorize all nondominated personal best solutions.
Keeping a personal archive yields
better results than traditional
methods.
25
26
A Particle Swarm Optimizer for
Constrained Numerical Optimization
Cagnina, Esquivel, Coello
Uses a single method to handle all
constraints
Results reported using standard test
functions.
27
Modelling Group-Foraging
Behaviour with Particle Swarms
Cecilia Di Chio, Riccardo Poli, and Paolo Di Chio
Uses a nature-inspired technique to
model a naturla problem: group
foraging.
Results encouraging.
28
An evolutive approach for the delineation of local
labour markets
Florez, Casado, Martinez-Bernabeu
Using a GA for delineating local labour
markets in Valencia, Spain.
Uses a bunch of operators:
mutation,crossover.
Better results than classical algorithms
29
That's all
Thank you and enjoy the session
30