IEEE Computational Intelligence Magazine - February 2023 - 21

This aim of this special issue is to promote research on
representation of solutions (and the dual perspective
of fitness landscape manipulation) in CI, with a
particular focus on biomedical and health-related
disciplines.
peer review process, four papers were
finally accepted for publication in IEEE
Computational IntelligenceMagazine.
The first paper, titled " Simplifying Fitness
Landscapes using Dilation Functions
evolved with Genetic Programming " ,is
based on the idea ofusing deformations of
the fitness landscape to simplify the optimization
problem by " expanding " and
" compressing " specific regions. The concept
behind Dilation Functions (DFs) is to
remap the parameters of candidate solutions
to perform a better exploration of
the most promising areas, while avoiding
regions characterized by worse fitness.
This task is clearly problem-dependent.
To have arbitrary DFs tailored on the fitness
optimized, the authors proposed a
novel approach based on Genetic Programming
named GP4DFs, which
evolves a specific DF for each dimension
of the search space. The performance of
GP4DFs was compared against a bilevel
optimization algorithm named GAFSTPSO,
which exploits an outer layer
based on a ðm þÞ genetic algorithm to
evolve the structure of the DFs, while an
inner layer uses the self-tuning swarm
intelligence algorithm FST-PSO to identify
the optimal parameters. The performance
ofthe two methods was compared
usingavarietyofbenchmark functions
(including the CEC'17 suite) ofup to 100
dimensions. According to the authors,
GP4DFs outperformed GA-FSTPSO in
the majority of the tests. Moreover,
GP4DFs seems to require a smaller budget
(in terms offitnessevaluations)thanGAFSTPSO
to achieve similar results.
The second paper, titled " Graph Lifelong
Learning: a survey " ,proposes a thorough
review of methods for the lifelong
learning of graph-structured data. As a
matter of fact, the vast majority of articles
on graph learning assumes a static representation,
in which the complete structure
ofthe graph is known since the beginning
of the training process. However, this is
not possible in the case of continuously
changing or expanding graphs, where
new instances and/or new classes might
emerge incrementally. The paper organizes
the discussions in four groups: architectural,
rehearsal, approach, and hybrid
approaches. The state-of-the-art is presented
and tested on various scenarios,
addressing the open issues oflifelong graph
learning, including catastrophic forgetting
that afflicts continual machine learning
algorithms. Covering more than one hundred
works on graph lifelong learning, this
survey presents a valuable starting point
for all researchers interested in this emerging
hot topic on the boundary between
lifelong learning and graph learning.
The third paper, titled " AMultiTransformation
Evolutionary Framework
for Influence Maximization in Social
Networks " , tackles the problem of finding
the " influencers " in a social network
in a more efficient way than the usual
one of modeling the spread of information
via a diffusion process, which requires
alarge number ofsimulationsviaaMonte
Carlo process. In particular, the task is to
find the optimal subset of nodes that are
able to spread the most information in
the network. The approach used by the
authorsistodefine a multi-transformation
evolutionary framework to avoid an a priori
selection of a proxy model where the
optimization is performed. The individuals
in the multiple transformations are not
independent but are instead used to perform
a knowledge transfer process. The
paper presents an improvement in solving
the problem of influence maximization
while using less computational resources
than the existing methods.
The fourth paper, titled " Self-supervised
Fusion for Multi-modal Medical
Images via Contrastive Auto-encoding
and Convolutional Information Exchange " ,
concerns the combination of
medical images with different methods
while preserving the complementary
information from different sources. This
method aims to help in the diagnoses
and treatments of patients. The authors
formulate the multi-modal image fusion
problem as a contribution estimation
problem, making use ofa convolutional
autoencoder - being able to merge local
and global features - and a multi-convolutional
information exchange network,
to allow the estimation of the
contribution of each source image. The
experimental results show the superiority
of the proposed approach compared
to the state-of-the-art, demonstrating a
clear improvement on the state of
multi-modal image fusion.
The guest editors ofthis special issue
would like to thank Professor ChuanKang
Ting, Editor-in-Chief of IEEE
Computational Intelligence Magazine,
for the great support in developing this
special issue. We also take the opportunity
to thank all the authors for submitting
their valuable research outcomes, as
well as the reviewers, who have critically
evaluated the papers.
Wewould like to thank Prof.Daniel
" Dan " Ashlock, who suddenly left
this world during the development
of this special issue. His visions and
contributions were fundamental to
develop this area ofresearch.
FEBRUARY 2023 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 21
IEEE Computational Intelligence Magazine - February 2023

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - February 2023
