Fairness is a crucial consideration in machine learning, yet class imbalance poses significant challenges to building fair models. While ensemble learning is widely used in addressing fairness under class imbalance, current methods face two critical limitations. First, the diversity among base models is not explicitly considered during training, which weakens the complementarity among base models. Second, because base models are updated on essentially the same data distribution, they may not encounter enough biased or challenging cases, so the training signal cannot focus on where unfair errors occur most frequently. To address these limitations, we propose fairer ensembles via co-evolution (FairCoEv), which co-evolves base models and data distribution. FairCoEv adaptively evolves the data distribution to sample more from regions that trigger unfair or incorrect predictions, ensuring the ensemble keeps learning from the most informative cases. Simultaneously, FairCoEv evolves base models by explicitly incorporating a diversity regularisation term based on negative correlation learning to foster complementarity alongside accuracy and fairness optimisation. Extensive experiments on 30 datasets with varying imbalance degrees, covering both binary and multi-class tasks as well as settings with single and multiple sensitive attributes, show that FairCoEv significantly improves fairness with only a minor loss in accuracy, achieving a better fairness–accuracy trade-off than state-of-the-art methods. Notably, on the unseen fairness metric overall accuracy equality, which is not directly optimised during training, our method achieves the best performance, ranking first among all compared methods.

Online content generation enables automatic and adaptive creation of diverse and playable game content for maximizing player experience or testing Artificial Intelligence (AI) algorithms. Multiple diversity metrics have been formulated on different content facets in the literature, while some of them conflict with one another. Existing work addresses this multi-dimensional diversity nature by converting those metrics into one term that is further used to direct the training of content generators. However, each generator is trained to meet the preference specified by the weights and fails to fully interpret the relationships among these metrics or provide different trade-offs. This paper proposes a multi-objective procedural content generation via reinforcement learning to train a set of generators that create diverse game content in an online manner while balancing the trade-off between multiple diversity metrics with playability as a constraint. Our framework is compared with state-of-the-art approaches on the commonly used Mario-AI benchmark. Results show that our framework is capable of increasing the diversity of the generator distribution while accelerating the convergence during the early stages of model training. Our approach enables researchers, designers, and practitioners to gain a better understanding of the relationship among conflicting diversity metrics, allowing them to generate content more efficiently and accurately tailored to specific needs.

Recent neural solvers have demonstrated promising performance in learning to solve routing problems. However, existing studies are primarily based on one-off training on one or a set of predefined problem distributions and scales, i.e., tasks. When a new task arises, they typically rely on either zero-shot generalization, which may be poor due to the discrepancies between the new task and the training task(s), or fine-tuning the pretrained solver on the new task, which possibly leads to catastrophic forgetting of knowledge acquired from previous tasks. This paper explores a novel lifelong learning paradigm for neural VRP solvers, where multiple tasks with diverse distributions and scales arise sequentially over time. Solvers are required to effectively and efficiently learn to solve new tasks while maintaining their performance on previously learned tasks. Consequently, a novel framework called Lifelong Learning Router with Behavior Consolidation (LLR-BC) is proposed. LLR-BC consolidates prior knowledge effectively by aligning behaviors of the solver trained on a new task with the buffered ones in a decision-seeking way. To encourage more focus on crucial experiences, LLR-BC assigns greater consolidated weights to decisions with lower confidence. Extensive experiments on capacitated vehicle routing problems and traveling salesman problems demonstrate LLR-BC's effectiveness in training high-performance neural solvers in a lifelong learning setting, addressing the catastrophic forgetting issue, maintaining their plasticity, and improving zero-shot generalization ability.

Autonomous Driving Systems (ADS) are safety-critical. Abundant and various driving scenarios are required to train accurate and robust models, and comprehensively test each module of autonomous driving systems (i.e., perception, tracking, prediction, planning, and control modules). However, collecting driving scenario data from the real physical world is expensive and inefficient. Most existing works generate simulated driving scenarios by varying the behaviors of dynamic objects on simple road networks (e.g., highways), while the influence of roadside structures and scenarios with complex road networks are not considered. This paper proposes a novel driving scenario generation approach, Automated Scenario Crafting (AutoSceCraft), to automatically produce abundant driving scenarios containing various road networks, traffic rules, roadside structures, and dynamic objects at low cost. To validate the effectiveness and efficiency of our proposed framework, AutoSceCraft is integrated into three popular driving simulators, including SMARTS, esmini, and CARLA. Numerical experiments and scenario visualization results show that AutoSceCraft can generate effectively and efficiently various driving scenarios from scratch for testing and training various modules (including perception, prediction, and planning modules) within autonomous driving systems.

Transfer learning-based dynamic multi-objective evolutionary algorithms (Tr-DMOEAs) represent one of the earliest feature-representation-transfer optimization approaches in dynamic multi-objective optimization. Our previous investigations into Tr-DMOEAs revealed transfer effectiveness on problems with drastic shifts in Pareto-optimal sets (PSs), and demonstrated the superiority of a linear kernel over the original Gaussian kernel. This paper conducts comprehensive comparative studies on knowledge transfer in Tr-DMOEA variants with a linear kernel. We firstly investigate how transfer works in Tr- DMOEA with a linear kernel. We then propose six new Tr- DMOEA variants with linear feature representations to study the impact of different components of knowledge transfer in Tr- DMOEA. Among them, two variants have different approaches to learning linear feature representations, and four variants use different source knowledge selection methods. We show that all these six variants have little impact on knowledge transfer in Tr-DMOEA (i.e., little impact on the dynamic optimization). Considering the computational cost of feature representation learning and its little impact on knowledge transfer in Tr- DMOEA, we then propose three other new Tr-DMOEA variants without learning feature representations to further investigate their impact on knowledge transfer. The three variants have different methods of selecting source knowledge, which are experimentally shown to have few impact on knowledge transfer in Tr-DMOEA either. Finally, we demonstrate the superiority of Tr-DMOEA variants without feature representation over those with it regarding solution quality and transfer efficiency. Our studies show that feature representation is not essential in transfer learning based dynamic optimization.

Existing neural solvers for vehicle routing problems (VRPs) are typically trained either in a one-off manner on a fixed set of pre-defined tasks or in a lifelong manner on several tasks arriving sequentially, assuming sufficient training on each task. Both settings overlook a common real-world property: problem patterns may drift continually over time, yielding massive tasks sequentially arising while offering only limited training resources per task. In this paper, we study a novel lifelong learning paradigm for neural VRP solvers under continually drifting tasks over learning time steps, where sufficient training for any given task at any time is not available. We propose Dual Replay with Experience Enhancement (DREE), a general framework to improve learning efficiency and mitigate catastrophic forgetting under such drift. Extensive experiments show that, under such continual drift, DREE effectively learns new tasks, preserves prior knowledge, improves generalization to unseen tasks, and can be applied to diverse existing neural solvers.

This survey comprehensively reviews the multi-dimensionality of game scenario diversity, spotlighting the innovative use of procedural content generation and other fields as cornerstones for enriching player experiences through diverse game scenarios. By traversing a wide array of disciplines, from affective modeling and multi-agent systems to psychological studies, our research underscores the importance of diverse game scenarios in gameplay and education. Through a taxonomy of diversity metrics and evaluation methods, we aim to bridge the current gaps in literature and practice, offering insights into effective strategies for measuring and integrating diversity in game scenarios. Our analysis highlights the necessity for a unified taxonomy to aid developers and researchers in crafting more engaging and varied game worlds. This survey not only charts a path for future research in diverse game scenarios but also serves as a handbook for industry practitioners seeking to leverage diversity as a key component of game design and development.

Multiobjective evolutionary learning (MOEL) has demonstrated its advantages of training fairer machine learning models considering a predefined set of conflicting objectives, including accuracy and different fairness measures. Recent works propose to construct a representative subset of fairness measures as optimisation objectives of MOEL throughout model training. However, the determination of a representative measure set relies on dataset, prior knowledge and requires substantial computational costs. What’s more, those representative measures may differ across different model training processes. Instead of using a static predefined set determined before model training, this paper proposes to dynamically and adaptively determine a representative measure set online during model training. The dynamically determined representative set is then used as optimising objectives of the MOEL framework and can vary with time. Extensive experimental results on 12 well-known benchmark datasets demonstrate that our proposed framework achieves outstanding performance compared to state-of-the-art approaches for mitigating unfairness in terms of accuracy as well as 25 fairness measures although only a few of them were dynamically selected and used as optimisation objectives. The results indicate the importance of setting optimisation objectives dynamically during training.

Meta-Black-Box Optimization (MetaBBO) garners attention due to its success in automating the configuration and generation of black-box optimizers, significantly reducing the human effort required for optimizer design and discovering optimizers with higher performance than classic human-designed optimizers. However, existing MetaBBO methods conduct one-off training under the assumption that a stationary problem distribution with extensive and representative training problem samples is pre-available. This assumption is often impractical in real-world scenarios, where diverse problems following shifting distribution continually arise. Consequently, there is a pressing need for methods that can continuously learn from new problems encountered on-the-fly and progressively enhance their capabilities. In this work, we explore a novel paradigm of lifelong learning in MetaBBO and introduce LiBOG, a novel approach designed to learn from sequentially encountered problems and generate high-performance optimizers for Black-Box Optimization (BBO). LiBOG consolidates knowledge both across tasks and within tasks to mitigate catastrophic forgetting. Extensive experiments demonstrate LiBOG's effectiveness in learning to generate high-performance optimizers in a lifelong learning manner, addressing catastrophic forgetting while maintaining plasticity to learn new tasks.

In recent years, the generation of diverse game levels has gained increasing interest, contributing to a richer and more engaging gaming experience. A number of level diversity metrics have been proposed in literature, which are naturally multi-dimensional, leading to conflicted, complementary, or both relationships among these dimensions. However, existing level generation approaches often fail to comprehensively assess diversity across those dimensions. This paper aims to expand horizons of level diversity by considering multi-dimensional diversity when training generative models. We formulate the model training as a multi-objective learning problem, where each diversity metric is treated as a distinct objective. Furthermore, a multi-objective evolutionary learning framework that optimises multiple diversity metrics simultaneously throughout the model training process is proposed. Our case study on the commonly used benchmark Super Mario Bros . demonstrates that our proposed framework can enhance multi-dimensional diversity and identify a Pareto front of generative models, which provides a range of tradeoffs among playability and two representative diversity metrics, including a content-based one and a player-centered one. Such capability enables decision-makers to make informed choices when selecting generators accommodating a variety of scenarios and the diverse needs of players and designers.

We introduce GVGAI-LLM, a video game benchmark for evaluating the reasoning and problem-solving capabilities of large language models (LLMs). Built on the General Video Game AI framework, it features a diverse collection of arcade-style games designed to test a model's ability to handle tasks that differ from most existing LLM benchmarks. The benchmark leverages a game description language that enables rapid creation of new games and levels, helping to prevent overfitting over time. Each game scene is represented by a compact set of ASCII characters, allowing for efficient processing by language models. GVGAI-LLM defines interpretable metrics, including the meaningful step ratio, step efficiency, and overall score, to assess model behavior. Through zero-shot evaluations across a broad set of games and levels with diverse challenges and skill depth, we reveal persistent limitations of LLMs in spatial reasoning and basic planning. Current models consistently exhibit spatial and logical errors, motivating structured prompting and spatial grounding techniques. While these interventions lead to partial improvements, the benchmark remains very far from solved. GVGAI-LLM provides a reproducible testbed for advancing research on language model capabilities, with a particular emphasis on agentic behavior and contextual reasoning.

Dynamic material handling (DMH) involves the assignment of dynamically arriving material transporting tasks to suitable vehicles in real time for minimizing makespan and tardiness. In real-world scenarios, historical task records are usually available, which enables the training of a decision policy on multiple instances consisting of historical records. Recently, reinforcement learning (RL) has been applied to solve DMH. Due to the occurrence of dynamic events such as new tasks, adaptability is highly required. Solving DMH is challenging since constraints, including task delay, should be satisfied. A feedback is received only when all tasks are served, which leads to sparse reward. Besides, making the best use of limited computational resources and historical records for training a robust policy is crucial. The time allocated to different problem instances would highly impact the learning process. To tackle those challenges, this article proposes a novel adaptive constrained evolutionary RL (ACERL) approach, which maintains a population of actors for diverse exploration. ACERL accesses each actor for tackling sparse rewards and constraint violation to restrict the behavior of the policy. Moreover, ACERL adaptively selects the most beneficial training instances for improving the policy. Extensive experiments on eight training and eight unseen test instances demonstrate the outstanding performance of ACERL compared with several state-of-the-art algorithms. Policies trained by ACERL can schedule the vehicles while fully satisfying the constraints. Additional experiments on 40 unseen noised instances show the robust performance of ACERL. Cross validation further presents the overall effectiveness of ACREL. Besides, a rigorous ablation study highlights the coordination and benefits of each ingredient of ACERL.

Many large-scale combinatorial optimization problems (LSCOPs) are challenging to solve due to their high dimensionalities and complex search spaces. To deal with tens of thousands of decision variables, problem reduction approaches have proven effective in reducing the original problem instance to a more manageable size, thereby decreasing the dimensionality of problem instances before optimization. However, since the dimensionality of transformed problem instances remains fixed during the optimization process, the effectiveness of existing methods is highly dependent on the selected features. In this paper, instead of only transforming the original problem instance once prior to the optimization, we propose a novel learning-based adaptive problem reduction (LAPR) framework to facilitate solving LSCOPs. Based on a learning model, the LAPR framework initially transforms the original problem instance into a low-dimensional one, and then dynamically increases the problem dimensionality when necessary during the optimization. To evaluate the effectiveness and efficiency of our framework, we have applied it to solve uncapacitated facility location problems (UFLP) using three effective features to construct a machine learning model learning from small instances. Based on this model, the LAPR framework first identifies a subset of the most promising facilities to construct an initial reduced problem instance through a novel optimal-k estimation strategy. Then, it incrementally incorporates potentially promising facilities into candidate solutions during the evolutionary process to improve the optimization performance. Extensive experimental studies on a series of large-scale UFLP benchmarks demonstrate that our proposed LAPR framework significantly enhances the performance of existing solution methods and consistently outperforms state-of-the-art problem reduction frameworks.

Traveling salesman problem (TSP) is a combinatorial optimization problem, serving as basis for many real-world applications (e.g., transportation planning, circuit board design, and DNA sequencing). In TSP, it is common to encounter some dynamic events, for example, traffic jam, and roadworks in transportation planning. To deal with such dynamic TSP (DTSP) scenarios, numerous techniques have been designed over decades. In this article, we take a different perspective to study DTSP. Instead of focusing only on algorithm design for DTSP, we investigate how dynamic events in DTSP affect its fitness landscape (e.g., the location of local optimal solutions and the ruggedness level of search space). We consider three dynamic events, including node addition, node deletion and weight changes, and analyze how they affect the TSP with respect to the overall landscape structure and solution optimality. Experimental results show that the weight change event has great effect on the problem's fitness landscape, introducing more local optima and reducing the basin of attraction for the global optimum. This may suggest that search algorithms need to have stronger exploration capability when handling weight change dynamic events. Furthermore, our experimental studies also demonstrate that the dynamic solution adaptation strategy on the original global optimum is effective for tracking the new optimum after dynamic changes.

Large language models (LLMs) have demonstrated tremendous potential in game playing, while little attention has been paid to their ethical implications in those contexts. This work investigates and analyses the ethical considerations of applying LLMs in game playing, using Werewolf, also known as Mafia, as a case study. Gender bias, which affects game fairness and player experience, has been observed from the behaviour of LLMs. Some roles, such as the Guard and Werewolf, are more sensitive than others to gender information, presented as a higher degree of behavioural change. We further examine scenarios in which gender information is implicitly conveyed through names, revealing that LLMs still exhibit discriminatory tendencies even in the absence of explicit gender labels. This research showcases the importance of developing fair and ethical LLMs. Beyond our research findings, we discuss the challenges and opportunities that lie ahead in this field, emphasising the need for diving deeper into the ethical implications of LLMs in gaming and other interactive domains.

Appropriate selection of search operators plays a critical role in meta-heuristic algorithm design. Adaptive selection of suitable operators to the characteristics of different optimization stages is an important task that owns promising potential to improve the performance of a meta-heuristic algorithm. A variety of adaptive operator selection methods have been proposed in last decades, from the machine learning and optimization communities. However, the existing studies have not been systematically reviewed so far. To fill the gap, this paper provides a comprehensive survey of adaptive operator selection for meta-heuristics. According to the information required for selection, adaptive operator selection methods are classified into two categories: (i) stateless methods and (ii) state-based methods. Each category is further summarized into several key components. The strategies of each component belonging to the two categories are reviewed respectively. The motivation, strengths and weaknesses of the proposed strategies are also discussed. Furthermore, studied meta-heuristics and optimization problems in the literature are summarized. The effects from the difference of meta-heuristics and problems to the specific design of methods are discussed, together with the guidance of selecting the suitable method in different application scenarios. At the end, emerging challenges that could guide further research are discussed.

Large Language Models (LLMs) have made significant strides in the field of artificial intelligence, showcasing their ability to interact with humans and influence human cognition through information dissemination. However, recent studies have brought to light instances of bias inherent within these LLMs, presenting a critical issue that demands attention. In our research, we delve deeper into the intricate challenge of harmonising accuracy and fairness in the enhancement of LLMs. While improving accuracy can indeed enhance overall LLM performance, it often occurs at the expense of fairness. Overemphasising optimisation of one metric invariably leads to a significant degradation of the other. This underscores the necessity of taking into account multiple considerations during the design and optimisation phases of LLMs. Therefore, we advocate for reformulating the LLM training process as a multi-objective learning task. Our investigation reveals that multi-objective evolutionary learning (MOEL) methodologies offer promising avenues for tackling this challenge. Our MOEL framework enables the simultaneous optimisation of both accuracy and fairness metrics, resulting in a Pareto-optimal set of LLMs. In summary, our study sheds valuable lights on the delicate equilibrium between accuracy and fairness within LLMs, which is increasingly significant for their real-world applications. By harnessing MOEL, we present a promising pathway towards fairer and more efficacious AI technologies.

Recently, the emergence of large language models (LLMs) has unlocked new opportunities for procedural content generation. However, recent attempts mainly focus on level generation for specific games with defined game rules such as Super Mario Bros. and Zelda. This paper investigates the game generation via LLMs. Based on video game description language, this paper proposes an LLM-based framework to generate game rules and levels simultaneously. Experiments demonstrate how the framework works with prompts considering different combinations of context. Our findings extend the current applications of LLMs and offer new insights for generating new games in the area of procedural content generation.

Games have been the perfect test-beds for artificial intelligence research for the characteristics that widely exist in real-world scenarios. Learning and optimisation, decision making in dynamic and uncertain environments, game theory, planning and scheduling, design and education are common research areas shared between games and real-world problems. Numerous open-source games or game-based environments have been implemented for studying artificial intelligence. In addition to single- or multi-player, collaborative or adversarial games, there has also been growing interest in implementing platforms for creative design in recent years. Those platforms provide ideal benchmarks for exploring and comparing artificial intelligence ideas and techniques. This paper reviews the games and game-based platforms for artificial intelligence research, provides guidance on matching particular types of artificial intelligence with suitable games for testing and matching particular needs in games with suitable artificial intelligence techniques, discusses the research trend induced by the evolution of those games and platforms, and gives an outlook.

Many real multi-objective optimization problems with 20-50 decision variables often have only a small number of function evaluations available, because of their heavy time/money burden. Therefore, surrogate models are often utilized as alternatives for expensive function evaluations. However, the approximation error of the surrogate model is inevitable compared to the real function evaluation. The approximation error has a similar impact on the algorithm as noise, i.e., different optimization stages suffer from various impacts. Therefore, the current optimization stage can be indirectly detected via measuring the impact of the noise formulated by the approximation error. In addition, the rising dimension of the search space leads to an increase in the approximation errors of the surrogate models, which poses a huge challenge for existing surrogate-assisted multi-objective evolutionary algorithms. In this work, we propose a stage-adaptive surrogate-assisted multi-objective evolutionary algorithm to solve the medium-scale optimization problems. In the proposed algorithm, the ensemble model consisting of the latest and historical models is used as the surrogate model, on the basis of which a set of potential candidates can be discovered. Then, a stage-adaptive infill sampling strategy selects the most suitable sampling strategy by analyzing the demand of the current optimization stage on convergence, diversity, model accuracy to sample from the candidates. As for the current optimization stage, it is detected by a noise impact indicator, where the approximation errors of surrogate models are formulated as noise. The experimental results on a series of medium-scale expensive test problems demonstrate the superiority of the proposed algorithm over six state-of-the-art compared algorithms.