IEEE Computational Intelligence Magazine - May 2020 - 21

Only three benchmark datasets (i.e., the
Recent research has made some progress on the
INRC2010 competition [56], Nottingham and
UK [55] datasets) have been widely tested in
generality of algorithms. However, reusability
NRP, the first two with extensive NRP variants.
of algorithms remains underexplored.
An interesting study on INRC2010 [52] focuses
on choosing a compact set of low-level heuristics. The methods developed showed to be highly effective on
components may be analyzed to derive new knowledge
the benchmark as well as two real-world scheduling problems.
potentially transferable to solve unseen p. GCOP may conSuch analysis can also be conducted automatically with the
tribute to addressing the challenging research issue of generaliGCOP model. In [22], it is found that simple configuration
ty and reusability of algorithms.
methods work as effective as complicated algorithms on effective
❏ Generality: Recent hyperheuristics have shown to be
low-level heuristics. Such studies provide useful insights on
able to address cross-domain COPs [58]. There is to our
knowledge, however, not yet a formal definition of algoa ! A 1.0 for further research in GCOP.
rithm generality in the literature. In [59], a new assessment
In this paper, we define the most basic a ! A 1.0 across
method has been proposed to evaluate hyperheuristics
COPs, aiming to establish the fundamentals of the GCOP
against four levels of generality, in terms of solving different
standard. More advanced research will be conducted as disproblem domains, problems, problem instances and benchcussed in Section IV, and also strongly encouraged from the
marks, respectively. The automatically generated new algoresearch communities, to further enhance the GCOP stanrithms with GCOP can be evaluated against these four
dard toward automated algorithm design. Following the
levels of generality for solving different p. Note that the
recommended good practice in OR [20], updates of extensions and resources on GCOP will be provided at a dediassessment of algorithm optimality is different from that of
cated GCOP website at https://sites.google.com/view/
algorithm generality. The latter may also measure the
general-cop.
robustness and speed, in addition to solution quality for
multiple problems/domains.
IV. Discussions and Future Directions
❏ Reusability: Recent research has made some progress on
reusing algorithms, although the main research focus may
As a fast emerging topic in computational intelligence, evolunot be exactly on reusability. For example, the automatically
tionary computation and optimization research, automated
selected algorithms on training instances [10], [11] could be
algorithm design has recently attracted increasing research
reused to solve testing instances of certain similar features.
attention. In this paper, GCOP is formally established as a new
In generation hyperheuristics [60], new heuristics can be
standard to define various search algorithms in one model, proautomatically generated by using genetic programming
viding the fundamentals and opening a number of potential
based on problem state features [61], [62], thus could be
new research directions in automated algorithm design.
potentially reusable for problems of similar features. HowevNew knowledge toward automated algorithm
er, the problem of code bloat may lead to the issues of readdesign: The new GCOP model provides a standard for systemability and interpretability [63].
atic analysis on the basic a of different behaviors in the optiFundamentals of GCOP: Advanced theoretical investigamized c. Some studies in hyperheuristics identified a compact
tions are needed to underpin the fundamentals of the new
subset of effective low-level heuristics and revealed synergy
GCOP model in operational research.
among them, enabling effective methods to be built [53]. In
[57], a runtime analysis on a selection hyperheuristic shows that
❏ Evaluation of GCOP: In solving GCOP, the objective funconline reinforcement learning for configuring operators may
tion can be extended with multiple objectives including
perform poorer than a fixed distribution of operators. These
generality, reusability and computational time. The new peranalyses could also be conducted within the consistent GCOP
formance measure in [59] can be adopted in the objective
model. The balanced intensification and diversification may be
function to measure different levels of generality. In GCOP,
modeled in c considering synergy among a. New findings on
instead of designing algorithms using human expertise, as
happens in most of the research, the time is spent on autonew effective algorithms with different categories of algorithmatically searching for or composing the optimal c ) for p.
mic components may also lead to new knowledge and deeper
understanding in algorithm design and introduce new effective
The trade-off between solving a number of p and the
algorithms to the literature.
increased computational time presents another interesting
Generality and reusability of algorithms: In GCOP,
research issue. The c for each p can be further evaluated in
the automatically designed new algorithms evolve to perform
F to assess its convergence, the number of operations and
well for solving different p, thus may cater for similar types of
number of fitness evaluations used, using different statistical
measures as shown in [64].
new p with a certain level of generality and reusability.
Recent research has made some progress on the generality of
❏ No Free Lunch Theorem (NFL): Another interesting
algorithms. However, the reusability of algorithms remains
research issue is how NFL applies in solving GCOP, that is
underexplored. With the GCOP standard, the optimized
to explore the scope of generality for the generated new c.

MAY 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

https://sites.google.com/view/general-cop https://sites.google.com/view/general-cop

IEEE Computational Intelligence Magazine - May 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - May 2020