IEEE Computational Intelligence Magazine - May 2022 - 64

D. EMT in Multi-Scenario Optimization
Imagine designing cars for various driving conditions, international
markets (e.g., Asian, American), types of use (e.g., taxi, family
car), or other scenarios. During design optimization, every scenario
could lead to different mathematical representations of the objective
functions, even if their physical interpretations remain the
same. For instance, let
SK
), (),, ( )],
then a general formulation of a multi-scenario multi-objective
optimization problem (MSMOP) may be stated as [128], [129]:
12 mi
max ff fi S}, ..
st
{( [ xx xx X.
ii if !! (7)
Here, mi is the number of objectives in the i-th scenario, and
X is a unified search space. A straightforward all-at-once
approach tackles Eq. (7) by fusing all the objective functions
together into a gigantic MaOP. This may however lead to tractability
issues and the return of solutions that do not belong to
the Pareto set of individual scenarios. Hence, the solving of
each scenario as a separate task was advocated in [129], with
post-hoc coordination between the tasks. Clearly, such a recipe
for MSMOPs is ideally suited for the implicit parallelism and
inter-task transfers of EMT, facilitating the discovery of solutions
that are skilled for multiple scenarios.
A real-world use-case of such multi-scenario optimization
was studied in [130], where EMT was applied to support intrahour
optimal power flow under rapid load variations. Multiple
scenarios were generated to accurately represent the variations
in power demand, and the MFEA was used to derive optimized
solutions for all scenarios in a proactive look-ahead
manner. The obtained solution set could then be used as
explicit setpoints to correctively control power generation-
thus improving overall operational economy.
V. Conclusions
Evolutionary multitasking (EMT) is an emerging paradigm for
jointly solving multiple tasks in a single optimization run. The
basic idea is to allow tasks to exchange information, transferring
evolved skills amongst one another to facilitate the efficient
discovery of high-quality solutions. A wealth of research
has been conducted in recent years to turn this idea into computational
algorithms.
The main aim of this paper is to draw the attention of optimization
practitioners to the many real-world use-cases of EMT
algorithms. To this end, several application-oriented explorations
of multitasking were reviewed in Section III. These were categorized
into half a dozen broad categories, enabling readers to
zoom in on applications of their choice. It is worth highlighting
that even at the time of this article's review, a number of applied
EMT works have appeared in the literature. Examples include an
EMT method crafted for multigroup collision-avoidance control
for UAV swarms [131] (which outperformed state-of-the-art
collision-avoidance models), a multiple surrogate assisted multitask
algorithm for production optimization in petroleum engineering
[132], among others [54], [133], [134].
Transcending specific application areas, Section IV provided
a set of recipes by which selected problem formulations of
64 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | MAY 2022
= 12 f {, ,, } be a set of scenarios,
general interest could be recast as EMT instances. This inexhaustive
list (which is expected to grow in the years to come)
reveals tricks by which the concept of MTO can be uniquely
leveraged, together with the implicit parallelism of EAs, to
tackle difficult optimization problems. The transformation of
multi-X optimization into MTO thus unpacks a new box of tools for
practitioners, going beyond the typical rendition of EMT for simultaneously
solving distinct problem instances.
It must however be acknowledged that there are still fundamental
challenges for EMT to reliably transition from concept to
demonstration to deployment. Recent methodological reviews
on EMT detail these issues [12], [13]. Within any application
domain, the success of EMT depends on its ability to consistently
deliver performance gains over state-of-the-art single-task algorithms/models.
Ensuring scalability to many-tasks and robustness
when faced with unrelated tasks is also vital to minimize/alleviate
performance degradation. Hence, theoretical research on what,
how, and when to transfer in EMT remains paramount, in conjunction
with the rigorous experimental comparison of new methods
against competitive multi- and single-task solvers in scenarios with
related, unrelated, and a mix of related and unrelated tasks.
Acknowledgement
Abhishek Gupta was supported by the A*STAR AI3 HTPO
seed grant C211118016 on Upside-Down Multi-Objective
Bayesian Optimization for Few-Shot Design. The work was
also supported in part by the Cyber-Physical Production System
Research Program, under the IAF-PP Grant A19C1a0018.
Yaqing Hou was supported by the National Natural Science
Foundation of China under Grant 61906032.
References
[1] J. Luo, A. Gupta, Y.-S. Ong, and Z. Wang, " Evolutionary optimization of expensive
multiobjective problems with co-sub-pareto front gaussian process surrogates, "
IEEE Trans. Cybern., vol. 49, no. 5, pp. 1708-1721, May 2018, doi: 10.1109/TCYB.
2018.2811761.
[2] M. Jiang, Z. Huang, L. Qiu, W. Huang, and G. G. Yen, " Transfer learning-based
dynamic multiobjective optimization algorithms, " IEEE Trans. Evol. Comput., vol. 22, no.
4, pp. 501-514, Aug. 2017, doi: 10.1109/TEVC.2017.2771451.
[3] R. Miikkulainen and S. Forrest, " A biological perspective on evolutionary computation, "
Nat. Machine Intell., vol. 3, no. 1, pp. 9-15, 2021, doi: 10.1038/s42256-020-00278-8.
[4] A. Gupta and Y.-S. Ong, " Back to the roots: Multi-x evolutionary computation, "
Cogn. Comput., vol. 11, no. 1, pp. 1-17, 2019, doi: 10.1007/s12559-018-9620-7.
[5] J. Yao et al., " Self-adaptive multifactorial evolutionary algorithm for multitasking production
optimization, " J. Petroleum Sci. Eng., vol. 205, p. 108,900, 2021.
[6] A. Gupta, Y.-S. Ong, and L. Feng, " Insights on transfer optimization: Because experience
is the best teacher, " IEEE Trans. Emerg. Topics Comput. Intell., vol. 2, no. 1, pp. 51-64,
Feb. 2018, doi: 10.1109/TETCI.2017.2769104.
[7] A. T. W. Min, A. Gupta, and Y.-S. Ong, " Generalizing transfer Bayesian optimization
to source-target heterogeneity, " IEEE Trans. Automat. Sci. Eng., vol. 18, no. 4, pp.
1754-1765, Oct. 2021, doi: 10.1109/TASE.2020.3017644.
[8] A. Gupta, Y. S. Ong, and L. Feng, " Multifactorial evolution: Toward evolutionary
multitasking, " IEEE Trans. Evol. Comput., vol. 20, no. 3, pp. 343-357, Jun. 2016, doi:
10.1109/TEVC.2015.2458037.
[9] G. Li, Q. Zhang, and W. Gao, " Multipopulation evolution framework for multifactorial
optimization, " in Proc. Genetic Evolutionary Comput. Conf. Companion, 2018, pp. 215-216.
[10] M.-Y. Cheng, A. Gupta, Y.-S. Ong, and Z.-W. Ni, " Coevolutionary multitasking for
concurrent global optimization: With case studies in complex engineering design, " Eng.
Appl. Artif. Intell., vol. 64, pp. 13-24, 2017, doi: 10.1016/j.engappai.2017.05.008.
[11] Q. Xu, N. Wang, L. Wang, W. Li, and Q. Sun, " Multi-task optimization and multitask
evolutionary computation in the past five years: A brief review, " Mathematics, vol. 9,
no. 8, p. 864, 2021, doi: 10.3390/math9080864.
[12] E. Osaba, A. D. Martinez, and J. Del Ser, " Evolutionary multitask optimization: a methodological
overview, challenges and future research directions, " 2021, arXiv:2102.02558.
[13] T. Wei, S. Wang, J. Zhong, D. Liu, and J. Zhang, " A review on evolutionary multitask
optimization: Trends and challenges, " IEEE Trans. Evol. Comput., early access, Dec.
30, 2021, doi: 10.1109/TEVC.2021.3139437.
IEEE Computational Intelligence Magazine - May 2022

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - May 2022
