Seeding the initial population of multi-objective evolutionary algorithms: A computational study
Graphical abstract
Introduction
In many real-world applications trade-offs between conflicting objectives play a crucial role. As an example, consider engineering a bridge, where one objective might be costs to build and another durability of the bridge. For such problems, we need specialized optimizers that determine the Pareto front of mutually non-dominated solutions. There are several established multi-objective evolutionary algorithms (MOEA) and many comparisons on various test functions. However, most of them start with random initial solutions.
If prior knowledge exists or can be generated at a low computational cost, good initial estimates may generate better solutions with faster convergence. These good initial estimates are often referred to as seeds, and the method of using good initial estimates is referred to as seeding. These botanical terms are used to express the possibility that good solutions for the environment can develop from these starting points. In practice, a good initial seeding can make problem solving approaches competitive that would otherwise be inferior.
For single-objective evolutionary algorithms, methods such as seeding have been studied for about two decades; see, e.g., [17], [20], [23], [26], [30], [41] for studies and examples (see [27] for a recent categorization). For example, the effects of seeding for the Traveling Salesman Problem (TSP) and the job-shop scheduling problem (JSSP) were investigated in [32]. The algorithms were seeded with known good solutions in the initial population, and it was found that the results were significantly improved on the TSP but not on the JSSP. To investigate the influence of seeding on the optimization, a varying percentage of seeding was used, ranging from 25 to 75%. Interestingly, it was also pointed out that a 100% seed is not necessarily very successful on either problems [28]. This is one of the very few reports that shows seeding can in some cases be beneficial to an optimization process, but not necessarily always is. In [21] a seeding technique for dynamic environments was investigated. There, the population was seeded when a change in the objective landscape arrived, aiming at a faster convergence to the new global optimum. Again, some of the investigated seeding approaches were more successful than others.
One of the very few studies that can be found on seeding techniques for MOEAs is the one performed by Hernandez-Diaz et al. [22]. There, seeds were created using gradient-based information. These were then fed into the algorithm called Non-Dominated Sorting Genetic Algorithm II (NSGA-II, [10]) and the quality was assessed on the benchmark family ZDT ([44], named after the authors Zitzler, Deb, and Thiele). The results indicate that the proposed approach can produce a significant reduction in the computational cost of the approach.
In general, seeding is not well documented for multi-objective problems, even for real-world problems. If seeding is done, then typically the approach is outlined and used with the comment that it worked in “preliminary experiments”—the reader is left in the dark on the design process behind the used seeding approach. This is quite striking as one expects that humans can construct a few solutions by hand, even if they do not represent the ranges of the objectives well. The least that one should be able to do is to reuse existing designs, and to modify these iteratively towards extremes. Nevertheless, even this manual seeding is rarely reported.
In this paper, we are going to investigate the effects of two structurally different seeding techniques for five algorithms on 48 multi-objective optimization (MOO) problems.
As seeding we use the weighted-sum method, where the trade-off preferences are specified by non-negative weights for each objective. Solutions to these weighted-sums of objectives can be found with an arbitrary classical single-objective evolutionary algorithm. In our experiments we use the algorithm Covariance Matrix Adaptation Evolution Strategy (CMA-ES, [18]). Details of the two studied weighting schemes are presented in Section 2.1.
There are different ways to measure the quality of the solutions. A recently very popular measure is the hypervolume indicator, which measures the volume of the objective space dominated by the set of solutions relative to a reference point [43]. Its disadvantage is its high computational complexity [4], [3] and the arbitrary choice of the reference point. We instead consider the mathematically well founded approximation constant. In fact, it is known that the worst-case approximation obtained by optimal hypervolume distributions is asymptotically equivalent to the best worst-case additive approximation constant achievable by all sets of the same size [6]. For a rigorous definition, see Section 2. This notion of multi-objective approximation was introduced by several authors [19], [15], [31], [35], [36] in the 80s and its theoretical properties have been extensively studied [9], [12], [33], [34], [37].
We use the jMetal framework [13] and its implementation of NSGA-II [10], Strength Pareto Evolutionary Algorithm (SPEA2, [45]), S-Metric Selection Evolutionary Multi-Objective Algorithm (SMS-EMOA, [14]), and Indicator Based Evolutionary Algorithm (IBEA, [42]). Additionally to these more classical MOEAs, we also study Approximation Guided Evolution (AGE, [7]), which aims at directly minimizing the approximation constant and has shown to perform very well for larger dimensions [38], [39], [40]. For each of these algorithms we compare their regular behavior after a certain number of iterations with their performance when initialized with a certain seeding.
We compare the aforementioned algorithms on four common families of benchmark functions. These are DTLZ ([11], named after the authors Deb, Thiele, Laumanns and Zitzler), LZ09 ([29], named after the authors Li and Zhang), WFG ([24], named after the authors’ research group Walking Fish Group) and ZDT [44]. While the last three families only contain two- and three-dimensional problems, DTLZ can be scaled to an arbitrary number of dimensions.
Section snippets
Preliminaries
We consider minimization problems with d objective functions, where d ≥ 2 holds. Each objective function , 1 ≤ i ≤ d, maps from the considered search space S into the real values. In order to simplify the presentation we only work with the dominance relation on the objective space and mention that this relation transfers to the corresponding elements of S.
For two points x = (x1, …, xd) and y = (y1, …, yd), with we define the following dominance relation:
Experimental setup
We use the jMetal framework [13], and our code for the seeding as well as all used seeds are available online.2 As test problems we used the benchmark families DTLZ [11], ZDT [44], LZ09 [29], and WFG [24], We used the functions DTLZ 1-4, each with 30 function variables and with d ∈ {2, 4, 6, 8} objective values/dimensions.
In order to investigate the benefits of seeding even in the long run, we limit the calculations of the algorithms to a
Experimental results
Our results are summarized in Table 1, Table 2. They compare the approximation constant achieved with CornersAndCentre seeding (Table 1) and LinearCombinations seeding (Table 2) with the same number of iterations without seeding. As the seeding itself requires a number of fitness function evaluations (104 for CornersAndCentre and 105 for LinearCombinations), we allocate the seeded algorithms fewer fitness function evaluations. This makes it harder for the seeded algorithms to outperform its
Conclusions
Seeding can result in a significant reduction of the computational cost and the number of fitness function evaluations needed. We observe that there is an advantage on many common real-valued fitness functions even if computing an initial seeding reduces the number of fitness function evaluations available for the MOEA. For some functions we observe a dramatic improvement in quality and needed runtime (e.g., DTLZ4 and the LZ09 family).
For practitioners, our results show that it can be
Acknowledgements
The research leading to these results has received funding from the Australian Research Council (ARC) under grant agreement DP140103400 and from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no 618091 (SAGE).
References (45)
- et al.
Approximating the volume of unions and intersections of high-dimensional geometric objects
Comput. Geom.: Theory Appl.
(2010) - et al.
Approximating the least hypervolume contributor: NP-hard in general, but fast in practice
Theor. Comput. Sci.
(2012) - et al.
Approximation quality of the hypervolume indicator
Artif. Intell.
(2013) - et al.
Linkage learning through probabilistic expression
Comput. Methods Appl. Mech. Eng.
(2000) - et al.
An empirical investigation of meta-heuristic and heuristic algorithms for a 2D packing problem
Eur. J. Oper. Res.
(2001) - et al.
A hybrid genetic algorithm for the design of water distribution networks
Eng. Appl. Artif. Intell.
(2005) A hybrid genetic algorithm for the open shop scheduling problem
Eur. J. Oper. Res.
(2000)- et al.
Efficiently computing succinct trade-off curves
Theor. Comput. Sci.
(2005) - et al.
Efficient optimization of many objectives by approximation-guided evolution
Eur. J. Oper. Res.
(2015) - et al.
A hybrid genetic algorithm for the fitting of models to electrochemical impedance data
J. Electroanal. Chem.
(2002)
Simulated Binary Crossover for Continuous Search Space. Technical report
Faster hypervolume-based search using Monte Carlo sampling
Parameterized average-case complexity of the hypervolume indicator
Approximation-guided evolutionary multi-objective optimization
Bicriterion single machine scheduling with resource dependent processing times
SIAM J. Optim.
How good is the Chord algorithm?
A fast and elitist multiobjective genetic algorithm: NSGA-II
IEEE Trans. Evolut. Comput.
Scalable test problems for evolutionary multiobjective optimization
Small approximate Pareto sets for biobjective shortest paths and other problems
SIAM J. Comput.
The jMetal framework for multi-objective optimization: design and architecture
An EMO algorithm using the hypervolume measure as selection criterion
Methods of numerical solution of multicriterion problem
Cited by (40)
Identifying vulnerabilities of industrial control systems using evolutionary multiobjective optimisation
2024, Computers and Security3D-flight route optimization for air-taxis in urban areas with Evolutionary Algorithms and GIS
2023, Journal of Air Transport ManagementBoosting ant colony optimization via solution prediction and machine learning
2022, Computers and Operations ResearchMulti-objective integer programming approaches to Next Release Problem — Enhancing exact methods for finding whole pareto front
2022, Information and Software TechnologyCitation Excerpt :The seeding mechanism is adopted and seeds are randomly selected non-dominated solutions found by the exact methods. Seeds are used in the progress during population initialization and offspring population creation [22,23]. Besides elapsed time and number of non-dominated solutions for each method finds, we adopt 3 other indicators for showing the quality of each method on all datasets.
Quantifying uncertainty in Pareto fronts arising from spatial data
2021, Environmental Modelling and SoftwareCitation Excerpt :Given that the single-objective optimal solutions of the multi-objective problem can be derived with a much lower computational effort, we can cost-effectively apply the exterior sampling method on the single-objective optima to compute the uncertainty in the single points at the outer ends of the Pareto fronts. To extend the available information from the single objectives to the uncertainty in the Pareto front, we execute the multi-objective optimization including the single objective optima, with a method called seeding (Friedrich and Wagner, 2015). Seeding entails that a part of the random initial solutions of the optimization is replaced by better solutions, in our case the single-objective optima.
A general framework and guidelines for benchmarking computational intelligence algorithms applied to forecasting problems derived from an application domain-oriented survey
2020, Applied Soft Computing JournalCitation Excerpt :GA finds nearly optimal solution via an iterative process of applying the genetic operators: selection, mutation, crossover to a population of individuals, minimizing the fitness function defined by the problem instance [16]. The population initialization can be done via random procedures or more informed ones as e.g. by following the guidelines for seeding the initial population of multi-objective evolutionary algorithms given in [17] or following the lessons learned in the study of GA optimization performance evaluation described in [18]. Genetic programming (GP) [19] was introduced by J. R. Koza in 1992 as a generalization of GA to populations of programs.