Multi-objective Optimization Problems (MOPs) where objectives have different levels of importance in decision-making, known as Mixed Pareto-Lexicographic MOPs with Priority Levels (PL-MPL-MOPs), are increasingly prevalent in real-world applications. General-purpose Multi-Objective Evolutionary Algorithms (MOEAs) that treat all objectives equally not only increase the workload of decision-making but also suffer from computational inefficiencies due to the necessity of generating many additional solutions. Conversely, strictly adhering to Priority Levels (PLs) during optimization can easily result in premature convergence within some PLs. To address this issue, we suggest an effective Balanced Adaptive Subspace Collaboration (BASC) method in this paper. Specifically, this method decomposes the search space into sub-fronts based on PLs and utilizes a sampling mechanism that operates exclusively within subspaces formed by sub-fronts at the same PL to generate new solutions, thereby emphasizing the exploitation of individual PLs. Furthermore, a set of parameters is employed to control the strictness of adherence to each PL, with these parameters adaptively adjusted to balance exploration across different PLs. The two mechanisms are then collaboratively integrated into MOEAs. Comprehensive experimental studies on benchmark problems and a set of complex job-shop scheduling problems in semiconductor manufacturing demonstrate the competitiveness of the proposed method over existing methods.

A series of surrogate-assisted evolutionary algorithms (SAEAs) have been proposed for expensive multi-objective optimization problems (EMOPs), building cheap surrogate models to replace the expensive real function evaluations. However, the search efficiency of these SAEAs is not yet satisfactory. More efforts are needed to further exploit useful information from the real function evaluations in order to better guide the search process. Facing this challenge, this paper proposes a Hyperbolic Neural Network (HNN) based preselection operator to accelerate the optimization process based on limited evaluated solutions. First, the preselection task is modeled as a multi-label classification problem where solutions are classified into different layers (ordinal categories) through -relaxed objective aggregation. Second, in order to resemble the hierarchical structure of candidate solutions, a hyperbolic neural network is applied to tackle the multi-label classification problem. The reason for using HNN is that hyperbolic spaces more closely resemble hierarchical structures than Euclidean spaces. Moreover, to alleviate the data deficiency issue, a data augmentation strategy is employed for training the HNN. In order to evaluate its performance, the proposed HNN-based preselection operator is embedded into two surrogate-assisted evolutionary algorithms. Experimental results on two benchmark test suites and three real-world problems with up to 11 objectives and 150 decision variables involving seven state-of-the-art algorithms demonstrate the effectiveness of the proposed method.

Messenger RNA (mRNA) vaccines have emerged as highly effective strategies in the prophylaxis and treatment of diseases. mRNA design, a key to the success of mRNA vaccines, in-volves finding optimal codons and increasing secondary structure stability to lengthen mRNA half-life, ultimately enhancing protein expression. Despite receiving widespread attention, most methods primarily rely on manual design, which is time-consuming and labor-intensive. While optimization approaches can alleviate this issue, existing methods still exhibit critical limitations caused by conflicts between codon usage and mRNA structural stability, compounded by the vast design space of mRNA resulting from the presence of synonymous codons. In this paper, a novel multi-objective evolutionary optimization-based mRNA design method is proposed. We first formulate the mRNA design problem as a multi-objective optimization problem and then develop an Elite Archive-Assisted Multi-Objective Evolutionary algorithm for mRNA Design, namely EAA-MOED, by incorporating a novel elite archive-assisted method into a weighted optimization framework to improve search efficiency. Experimental studies, involving two state-of-the-art mRNA design methods and five well-known MOEAs, show the competitiveness of the proposed EAA-MOED in mRNA design.

The field of adversarial textual attack has significantly grown over the past few years, where the commonly considered objective is to craft adversarial examples (AEs) that can successfully fool the target model. However, the imperceptibility of attacks, which is also essential for practical attackers, is often left out by previous studies. In consequence, the crafted AEs tend to have obvious structural and semantic differences from the original human-written text, making them easily perceptible. In this work, we advocate leveraging multi-objectivization to address such an issue. Specifically, we reformulate the problem of crafting AEs as a multi-objective optimization problem, where the attack imperceptibility is considered as an auxiliary objective. Then, we propose a simple yet effective evolutionary algorithm, dubbed HydraText, to solve this problem. HydraText can be effectively applied to both score-based and decision-based attack settings. Exhaustive experiments involving 44,237 instances demonstrate that HydraText consistently achieves competitive attack success rates and better attack imperceptibility than the recently proposed attack approaches. A human evaluation study also shows that the AEs crafted by HydraText are more indistinguishable from human-written text. Finally, these AEs exhibit good transferability and can bring notable robustness improvement to the target model by adversarial training.

Due to computationally and/or financially costly evaluation, tackling expensive multi-objective optimization problems is quite challenging for evolutionary algorithms. One popular approach to these problems is building cheap surrogate models to replace the expensive real function evaluations. To this end, various kinds of surrogate-assisted evolutionary algorithms (SAEAs) have been proposed, building surrogate models which predict the fitness values, classifications, or relation of the candidate solutions. However, off-spring generation, despite its important role in evolutionary optimization, has not received enough attention in these SAEAs. In this paper, a regularity model based framework, namely RM-SAEA, is proposed for better offspring generation in expensive multi-objective optimization. To be specific, RM-SAEA is featured with a heterogeneous offspring generation module, which is composed of a regularity model and a general genetic operator. Moreover, in order to alleviate the data deficiency issue in the expensive optimization scenario, a data augmentation strategy is employed while training the regularity model. Finally, two representative SAEAs are embedded into RM-SAEA in order to instantiate the proposed framework. Experimental results on benchmark multi-objective problems with up to 10 objectives demonstrate that RM-SAEA achieves the best overall performance compared with 6 state-of-the-art algorithms.

Multi-fidelity simulation is an effective approach to balancing speed and accuracy in expensive simulation, and its performance is affected by the quality of multi-fidelity simulation models. Building high-quality simulation models is non-trivial, especially for complex systems, because current manual modeling methods require sufficient domain knowledge and experience, increasing the labor and time costs. Motivated by the issues, this paper focuses on one of the most crucial simulation types, discrete event simulation, and develops a computer-aid multi-fidelity simulation modeling method called Optimization-based Multi-fidelity Simulation Modeling (OMFSM). OMFSM formulates multi-fidelity simulation modeling as a bi-objective simulation optimization problem to optimize speed and accuracy. An efficient optimization algorithm called Multi-objective Simulation Optimization based on Hypervolume (MOSO-HV) is tailored to select a set of high-quality models. Experimental results in a digital twin emergency department demonstrate that the computer-aid modeling method builds more and better multi-fidelity simulation models than manual modeling and reveal the effectiveness of MOSO-HV for OMFSM. The utility of OMFSM in multi-fidelity simulation is also justified by a real-world optimization problem. Note to Practitioners—Multi-fidelity simulation is an essential technique to fulfill the demand for accuracy analysis and quick decision-making in Industrial 4.0, such as digital twins and virtual reality. The quality of multi-fidelity simulation models significantly influences the performance of multi-fidelity simulation. Developers currently build multi-fidelity simulation models manually, and their experience determines the model’s quality. To reduce the labor and time costs in constructing high-quality multi-fidelity simulation models, we propose a computer-aid method named OMFSM from optimization for the first time. Experiments on a real case prove that OMFSM lightens the burden of manual modeling and provides more and better models for multi-fidelity simulation.

The main target of multiobjective evolutionary algorithms (MOEAs) is to find a set of evenly distributed nondominated solutions that approximate the Pareto front (PF) of a multiobjective optimization problem (MOP). This means that the approximated set should be as close to the PF as possible, and as diverse as possible. The former is usually called a convergence criterion and the latter is called a diversity criterion. A variety of strategies have been proposed to meet the two criteria. However, as far as the diversity criterion is concerned, it is still a challenge to achieve an evenly distributed approximation set with different sizes for a problem with a complicated PF shape. To deal with this challenge, we propose a local subset selection (LSS) -based environmental selection for evolutionary multiobjective optimization in this article. LSS considers the environmental selection as a subset selection problem by choosing promising solutions from the combination of the parent and offspring populations. In LSS, a potential energy function is utilized as the objective function, which provides a heavy selection pressure on diversity as well as has low computational complexity. Furthermore, to balance search efficiency and quality, a local search strategy is used in LSS to make full use of objective information for acceleration. The proposed LSS strategy is embedded into some state-of-the-art Pareto-domination-based MOEAs, and the experimental results suggest that LSS can produce shape-invariant and evenly distributed nondominated sets with different population sizes.

Neural network pruning is a popular model compression method which can significantly reduce the computing cost with negligible loss of accuracy. Recently, filters are often pruned directly by designing proper criteria or using auxiliary modules to measure their importance, which, however, requires expertise and trial-and-error. Due to the advantage of automation, pruning by evolutionary algorithms (EAs) has attracted much attention, but the performance is limited for deep neural networks as the search space can be quite large. In this paper, we propose a new filter pruning algorithm CCEP by cooperative coevolution, which prunes the filters in each layer by EAs separately. That is, CCEP reduces the pruning space by a divide-and-conquer strategy. The experiments show that CCEP can achieve a competitive performance with the state-of-the-art pruning methods, e.g., prune ResNet56 for 63.42% FLOPs on CIFAR10 with -0.24% accuracy drop, and ResNet50 for 44.56% FLOPs on ImageNet with 0.07% accuracy drop.

Minimum cost seed selection for competitive influence maximization, which selects a set of key users (called seed set) to spread its influence widely into the network at a minimum cost in a competitive social network, is a key algorithmic problem in social influence analysis. Due to its application potential in multiple fields, such as market expansion, election campaigns, and cultural competition, numerous studies have been emerging recently. Despite these efforts, this problem has not been satisfactorily solved since not only finding a (nearly) optimal solution for cost minimization but also evaluating a seed set is computationally complex. Existing works either trade approximation guarantees for practical efficiency using heuristics, or vice versa due to costly Monte Carlo simulations. In this article, a competitive reverse influence estimation-based greedy (CRIEG) algorithm, which provides bounded approximation guarantees, but offers significantly improved empirical efficiency under the competitive independent cascade model, is proposed. The core of the algorithm is a novel estimation method that improves the efficiency by constructing representative sketches to avoid heavy repeated simulations without compromising its performance guarantees. The experimental results on eight real-world networks with up to 1.13 million users show that compared with state-of-the-art algorithms, our algorithm is the most efficient while keeping the best performance, and can be orders of magnitude faster.

Deep Reinforcement Learning (DRL) is an artificial intelligence technology that can complete decision-making tasks by interaction. It has been successfully applied to various games. However, there are still many challenges when this technique is applied to the industrial process control due to the low sample efficiency and the inability to deal with large time delay. In this paper, a novel Model Predictive Control (MPC) guided Reinforcement Learning Control (MP-RLC) scheme is proposed for the process control. In this scheme, Model predictive control is directly combined with Reinforcement Learning (RL) to guide the training process, thus greatly improving the sample efficiency of reinforcement learning and effectively solving the problem of time delay. The simulation results on both a third-order linear system and a nonlinear continuous stirred tank reactor (CSTR) system with large time delay demonstrate that this scheme can not only accelerate the training process but also improve the control performance, which is superior to both standalone RL and MPC schemes. The proposed approach may help to pave the way for DRL applied to industrial processes.

The performance of traditional multiobjective evolutionary algorithms (MOEAs) often deteriorates rapidly as the number of decision variables increases. While some efforts were made to design new algorithms by adapting existing techniques to large-scale single-objective optimization to the MOEA context, the specific difficulties that may arise from large-scale multiobjective optimization have rarely been studied. In this paper, the exclusive challenges along with the increase of the number of variables of a multiobjective optimization problem (MOP) are examined empirically, and the popular benchmarks are categorized into three groups accordingly. Problems in the first category only require MOEAs to have stronger convergence, and can thus be mitigated using techniques employed in large-scale single-objective optimization. Problems that require MOEAs to have stronger diversification but ignore a correlation between position and distance functions are grouped as the second. The rest of the problems that pose a great challenge to the balance between diversification and convergence by considering a correlation between position and distance functions are grouped as the third. While existing large-scale MOEAs perform well on the problems in the first two categories, they suffer a significant loss when applied to those in the third category. To solve large-scale MOPs in this category, we have developed a novel indicator-based algorithm with an enhanced diversification mechanism. The proposed algorithm incorporates a new solution generator with an external archive, thus forcing the search toward different subregions of the Pareto front using a dual local search mechanism. The results obtained by applying the proposed algorithm to a wide variety of problems (108 instances in total) with up to 8192 variables demonstrate that it outperforms eight state-of-the-art approaches on the examined problems in the third category and show its advantage in the balance between diversification and convergence.

Many real-world problems often have several, usually conflicting objectives. Traditional multi-objective optimization problems (MOPs) usually search for the Pareto-optimal solutions for this predicament. A special class of MOPs, the convex hull maximization problems which prefer solutions on the convex hull, has posed a new challenge for existing approaches for solving traditional MOPs, as a solution on the Pareto front is not necessarily a good solution for convex hull maximization. In this work, the difference between traditional MOPs and the convex hull maximization problems is discussed and a new Evolutionary Convex Hull Maximization Algorithm (ECHMA) is proposed to solve the convex hull maximization problems. Specifically, a Convex Hull-based sorting with Convex Hull of Individual Minima (CH-CHIM-sorting) is introduced, as well as a novel selection scheme, Extreme Area Extract-based selection (EAE-selection). Experimental results show that ECHMA significantly outperforms the existing approaches for convex hull maximization and evolutionary multi-objective optimization approaches in achieving a better approximation to the convex hull more stably and with a more uniformly distributed set of solutions.