Fairness in machine learning (ML) has garnered significant attention. However, current research has mainly concentrated on the distributive fairness of ML models, with limited focus on another dimension of fairness, i.e., procedural fairness. In this paper, we first define the procedural fairness of ML models by drawing from the established understanding of procedural fairness in philosophy and psychology fields, and then give formal definitions of individual and group procedural fairness. Based on the proposed definition, we further propose a novel metric to evaluate the group procedural fairness of ML models, called GPFFAE, which utilizes a widely used explainable artificial intelligence technique, namely feature attribution explanation (FAE), to capture the decision process of ML models. We validate the effectiveness of GPFFAE on a synthetic dataset and eight real-world datasets. Our experimental studies have revealed the relationship between procedural and distributive fairness of ML models. After validating the proposed metric for assessing the procedural fairness of ML models, we then propose a method for identifying the features that lead to the procedural unfairness of the model and propose two methods to improve procedural fairness based on the identified unfair features. Our experimental results demonstrate that we can accurately identify the features that lead to procedural unfairness in the ML model, and both of our proposed methods can significantly improve procedural fairness while also improving distributive fairness, with a slight sacrifice on the model performance.

Workflow scheduling has been extensively studied in distributed computing, with most research primarily focusing on single workflow scheduling problems. However, in real-world scenarios, multiple workflows from different individual users often need to be scheduled concurrently on shared computing resources, which raises significant fairness concerns among these workflows. Existing approaches typically overlook fairness in multiple workflow scheduling, leading to disproportionate completion time slowdowns across different workflows. To address this challenge, we introduce a novel fairness metric that quantitatively captures the slowdown disparity among multiple workflows and propose the FairFlowSR (Fair-Flow Stochastic Ranking) algorithm to ensure fairness among concurrent workflows. The FairFlowSR algorithm integrates two key components: a fair selection strategy that balances exploitation and exploration, and a stochastic ranking method that effectively handles fairness constraints. Extensive experimental results demonstrate that FairFlowSR significantly outperforms state-of-the-art algorithms, achieving superior fairness maintenance and competitive makespan optimization. These results validate the effectiveness of our approach in achieving a balanced trade-off between efficiency and fairness in multiple-workflow scheduling scenarios.

Explainable artificial intelligence (XAI) has gained significant attention, especially in AI-powered autonomous and adaptive systems (AASs). However, a discernible disconnect exists among research efforts across different communities. The machine learning community often overlooks “explaining to whom,” while the human-computer interaction community has examined various stakeholders with diverse explanation needs without addressing which XAI methods meet these requirements. Currently, no clear guidance exists on which XAI methods suit which specific stakeholders and their distinct needs. This hinders the achievement of the goal of XAI: providing human users with understandable interpretations. To bridge this gap, this article presents a comprehensive XAI roadmap. Based on an extensive literature review, the roadmap summarizes different stakeholders, their explanation needs at different stages of the AI system lifecycle, the questions they may pose, and existing XAI methods. Then, by utilizing stakeholders’ inquiries as a conduit, the roadmap connects their needs to prevailing XAI methods, providing a guideline to assist researchers and practitioners to determine more easily which XAI methodologies can meet the specific needs of stakeholders in AASs. Finally, the roadmap discusses the limitations of existing XAI methods and outlines directions for future research.

The feature attribution-based explanation (FAE) methods, which indicate how much each input feature contributes to the model’s output for a given data point, are one of the most popular categories of explainable machine learning techniques. Although various metrics have been proposed to evaluate the explanation quality, no single metric could capture different aspects of the explanations. Different conclusions might be drawn using different metrics. Moreover, during the processes of generating explanations, existing FAE methods either do not consider any evaluation metric or only consider the faithfulness of the explanation, failing to consider multiple metrics simultaneously. To address this issue, we formulate the problem of creating FAE explainable models as a multi-objective learning problem that considers multiple explanation quality metrics simultaneously. We first reveal conflicts between various explanation quality metrics, including faithfulness, sensitivity, and complexity. Then, we define the considered multi-objective explanation problem and propose a multi-objective feature attribution explanation (MOFAE) framework to address this newly defined problem. Subsequently, we instantiate the framework by simultaneously considering the explanation’s faithfulness, sensitivity, and complexity. Experimental results comparing with six state-of-the-art FAE methods on eight datasets demonstrate that our method can optimize multiple conflicting metrics simultaneously and can provide explanations with higher faithfulness, lower sensitivity, and lower complexity than the compared methods. Moreover, the results have shown that our method has better diversity, i.e., it provides various explanations that achieve different tradeoffs between multiple conflicting explanation quality metrics. Therefore, it can provide tailored explanations to different stakeholders based on their specific requirements.

With growing concerns about the unwanted bias or discrimination in machine learning, a number of fairness-aware machine learning algorithms have been developed to mitigate bias in the prediction. Because the objectives of accuracy and fairness are antagonistic, it is hard to balance the trade-off between them. Recently, Multi-objective Evolutionary Learning (MOEL) framework has been proposed to train a set of Pareto models with the consideration of accuracy and fairness simultaneously. However, this framework prefers to train more accurate models rather than fairer models due to the lack of gradient in terms of fairness. In this paper, MOEL is enhanced through introducing Adversarial Learning (AL). The MOEL-AL framework aims to maximize a set of predictors' ability to predict true labels and minimize the ability of an adversarial network to predict the sensitive attributes from the predictors' output. Specifically, the adversarial network can be regarded as a proxy of the undifferentiable fairness metrics, so it is possible to propagate gradients in terms of both accuracy and fairness for the predictors during the back-propagation process. Besides, the adversarial strength for different predictors is adjusted dynamically according to their fairness metric. Compared with the state-of-the-art methods, experimental studies on seven well-known datasets show that our method can provide a set of fairer Pareto models with little drop on accuracy.

Detecting whether a distribution shift has occurred in the dataset is a critical aspect when implementing machine learning models, as even a small shift in the data distribution may largely affect the performance of a machine learning model and thus cause the deployed model to fail. In this work, we focus on detecting harmful dataset shifts, i.e., shifts that are detrimental to the performance of the machine learning model. The existing methods usually detect whether there is a shift between two datasets according to the following framework: first carrying out dimensionality reduction on the datasets, then determining whether dataset shift exists according to the two-sample statistical test(s) on the reduced datasets. The knowledge contained in the model trained on the dataset is not utilized in the above described dataset shift detection framework. To address this, this paper proposes to take advantage of explainable artificial intelligence (XAI) techniques to exploit the knowledge in trained models when detecting harmful dataset shifts. Specifically, we employ the feature attribution explanation (FAE) method to capture the knowledge in the model and combine it with a widely-used two-sample test method, i.e., maximum mean difference (MMD), to detect harmful dataset shifts. The experimental results on more than twenty different shifts in three widely used image datasets demonstrate that the proposed method is more effective in identifying harmful dataset shifts than existing methods. Moreover, experiments on several different models show that the method is robust and effective over different models, i.e., its detection performance is not sensitive to the model used.

The extensive use of artificial intelligence (AI) in the real world brings some potential risks due to the undesirable behavior exhibited by AI systems using data-driven machine learning (ML) at their cores. Thus, preventing undesirable behaviors of ML, such as opacity (lack of transparency and explainability), unfairness (bias or discrimination), unsafety and insecurity, privacy disclosure, etc., is an imperative and pressing challenge. This work proposes an evolutionary constrained learning (ECL) framework for constructing ML models that can satisfy behavioral constraints so that the undesirable behaviors can be prevented. To evaluate our framework, we use it to create neural network models that preclude the undesirable behavior (that is, unfairness) on different benchmark datasets. The experimental results demonstrate the effectiveness of our proposed ECL approach for preventing undesirable behaviors of ML.

Search operator design and parameter tuning are essential parts of algorithm design. However, they often involve trial-and-error and are very time-consuming. A new differential evolution (DE) algorithm with adaptive exploration and exploitation control (AEEC-DE) is proposed in this work to tackle this challenge. The proposed method improves the performance of DE by automatically selecting trial vector generation strategies (both mutation and crossover operators) and dynamically generating the associated control parameter values. A probability-based exploration and exploitation measurement is introduced to estimate whether the state of each newly generated individual is in exploration or exploitation. The state of historical individuals is used to assess the exploration and exploitation capabilities of different generation strategies and parameter values. Then, the strategies and parameters of DE are adapted following the common belief that evolutionary algorithms (EAs) should start with exploration and then gradually change into exploitation. The performance of AEEC-DE is evaluated through experimental studies on a set of test problems and compared with several state-of-the-art adaptive DE variants.

Surrogate-assisted evolutionary algorithms (SAEAs), which use efficient surrogate models or meta-models to approximate the fitness function in evolutionary algorithms (EAs), are effective and popular methods for solving computationally expensive optimization problems. During the past decades, a number of SAEAs have been proposed by combining different surrogate models and EAs. This paper dedicates to providing a more systematical review and comprehensive empirical study of surrogate models used in single-objective SAEAs. A new taxonomy of surrogate models in SAEAs for single-objective optimization is introduced in this paper. Surrogate models are classified into two major categories: absolute fitness models, which directly approximate the fitness function values of candidate solutions, and relative fitness models, which estimates the relative rank or preference of candidates rather than their fitness values. Then, the characteristics of different models are analyzed and compared by conducting a series of experiments in terms of time complexity (execution time), model accuracy, parameter influence, and the overall performance when used in EAs. The empirical results are helpful for researchers to select suitable surrogate models when designing SAEAs. Open research questions and future work are discussed at the end of the paper.

The Capacitated Arc Routing Problem (CARP) is a general and challenging arc routing problem. As the problem size increasing, exact methods are not applicable, and heuristic and meta-heuristic algorithms are promising approaches to solve it. To obtain good performance, parameter values of heuristics or meta-heuristics should be properly set. In recent years, automatic parameter tuning, which includes off-line and online parameter tuning, has attracted considerable attention in the evolutionary computation community. At present, parameters are usually determined through simple off-line parameter tuning, such as empirical analysis or grid search, when designing algorithms for CARP. However, using off-line parameter tuning on CARP has some disadvantages, among which the computational cost is the serious one. This work proposed an online parameter tuning approach using exponential recency-weighted kernel density estimation (ERW-KDE), and combines it with the SAHiD algorithm, which is an hierarchical decomposition based algorithm for CARP, to constitute the online parameter tuned SAHiD (OPT-SAHiD) algorithm. The experimental results show that OPT-SAHiD significantly outperforms the compared algorithms on two CARP benchmark sets owing to the proposed online automatic parameter tuning approach. The proposed online automatic parameter tuning approach based on ERW-KDE not only improves the performance of SAHiD algorithm, but also removes the additional computational overhead required for offline parameter tuning.

Parameter tuning, that is, to find appropriate parameter settings (or configurations) of algorithms so that their performance is optimized, is an important task in the development and application of metaheuristics. Automating this task, i.e., developing algorithmic procedure to address parameter tuning task, is highly desired and has attracted significant attention from the researchers and practitioners. During last two decades, many automatic parameter tuning approaches have been proposed. This paper presents a comprehensive survey of automatic parameter tuning methods for metaheuristics. A new classification (or taxonomy) of automatic parameter tuning methods is introduced according to the structure of tuning methods. The existing automatic parameter tuning approaches are consequently classified into three categories: 1) simple generate-evaluate methods; 2) iterative generate-evaluate methods; and 3) high-level generate-evaluate methods. Then, these three categories of tuning methods are reviewed in sequence. In addition to the description of each tuning method, its main strengths and weaknesses are discussed, which is helpful for new researchers or practitioners to select appropriate tuning methods to use. Furthermore, some challenges and directions of this field are pointed out for further research.

The Capacitated Arc Routing Problem (CARP) is an essential and challenging problem in smart logistics. Parameter tuning is commonly encountered in designing and applying heuristic or meta-heuristic algorithms for CARP. Recently, automatic parameter tuning or hyper-parameter optimization, which focuses on automatically finding an optimal parameter setting of an algorithm for problems at hand, has attracted considerable attention and become popular for addressing parameter tuning problems. This paper studies automatic parameter tuning for advanced algorithms in solving CARP. When designing algorithms for CARP, parameters are usually determined through empirical analysis or following some rules of thumb. This paper uses an automatic parameter tuning approach, that is, Bayesian optimization method, to tune an algorithm called SAHiD, which is a scalable approach based on hierarchical decomposition for large-scale CARP. The experimental results show that the algorithm's performance can be significantly improved with automatic parameter tuning. The tuned SAHiD algorithm obtains better solutions and faster convergence speed than original SAHiD on test CARP instances.

Very expensive problems are very common in practical system that one fitness evaluation costs several hours or even days. Surrogate assisted evolutionary algorithms (SAEAs) have been widely used to solve this crucial problem in the past decades. However, most studied SAEAs focus on solving problems with a budget of at least ten times of the dimension of problems which is unacceptable in many very expensive real-world problems. In this paper, we employ Voronoi diagram to boost the performance of SAEAs and propose a novel framework named Voronoi-based efficient surrogate assisted evolutionary algorithm (VESAEA) for very expensive problems, in which the optimization budget, in terms of fitness evaluations, is only 5 times of the problem's dimension. In the proposed framework, the Voronoi diagram divides the whole search space into several subspace and then the local search is operated in some potentially better subspace. Additionally, in order to trade off the exploration and exploitation, the framework involves a global search stage developed by combining leave-one-out cross-validation and radial basis function surrogate model. A performance selector is designed to switch the search dynamically and automatically between the global and local search stages. The empirical results on a variety of benchmark problems demonstrate that the proposed framework significantly outperforms several state-of-art algorithms with extremely limited fitness evaluations. Besides, the efficacy of Voronoi-diagram is furtherly analyzed, and the results show its potential to optimize very expensive problems.