The flexible job shop scheduling problem is a complex NP-hard challenge critical to modern manufacturing systems, but existing research predominantly focuses on low-dimensional optimization, practical scenarios often require balancing high-dimensional conflicting objectives such as energy consumption, machine load, tardiness, and completion time. Current -objective algorithms struggle to maintain convergence and diversity in such high-dimensional spaces, leading to suboptimal Pareto fronts. To address this problem, we propose an improved NSGA-III (WDHS-NSGA-III) integrating a reference vector-guided subpopulation partitioning mechanism and a density-aware hybrid search strategy. The algorithm divides populations into subspaces using reference vectors to enable adaptive cross-subpopulation exploration, then applies localized neighborhood searches with five operators to prioritize under-explored regions. Experiments on benchmark instances demonstrate that WDHS-NSGA-III significantly outperforms commonly used methods, achieving balanced optimization across most objectives. The results validate its robustness in resolving high-dimensional tradeoffs, offering a valid solution for smart manufacturing scheduling.
Coral reefs are vital marine ecosystems that play a crucial role in supporting biodiversity, providing coastal protection, and driving global fisheries and tourism economies. However, these ecosystems are under increasing threat from coral bleaching, driven by climate change and human-induced stressors. Monitoring and mitigating coral bleaching require scalable, automated systems capable of accurate detection and classification. This study presents a synthesis of state-of-the-art deep learning models, including VGG16 and YOLO, for coral health assessment. Additionally, a novel annotated coral bleaching database is introduced, designed from diverse video and image sources to train machine learning models effectively. Our results demonstrate that these approaches not only improve the accuracy of coral health classification but also enhance the scalability and reliability of reef monitoring efforts, supporting marine conservation initiatives. Future work focuses on expanding datasets and exploring advanced architectures such as ResNet50 and EfficientNet to improve generalization and robustness.
This paper identifies and presents accelerated and automated AI solutions for major knowledge management problems in complex systems, especially in defense. From the interviews and surveys conducted on defense personnel and using defending-against-knowledge-loss triangulation, the study points to the following enduring issues: paperwork, siloed knowledge, security measures, and knowledge obsolescence due to employee turnover. The paper uses thematic analysis using the MAXQDA qualitative data analysis to elucidate the identified inefficiencies as barriers to decision-making and collaborative work that threaten organizational outcomes. In response to these issues, the study seeks to develop a model offering new solutions for the mechanised adoption of knowledge management processes, including knowledge capture, distribution, and enhancement of interoperability to support structured knowledge sharing. This model incorporates AI functionalities to automate processes and minimise paperwork to reduce decision-making latency and advance cross-departmental knowledge flow with the added advantage of accessing real-time information while conforming to the security requirements through operational access privileges.
Machine learning algorithms have been widely adopted across a variety of application domains. However, training models and evaluating their performance is often time-consuming and computationally expensive. Furthermore, the identification of (near-)optimal learning strategies, including design choices and hyperparameter configurations, remains a significant challenge, as these decisions substantially influence the quality of learning outcomes. To address these issues, Automated Machine Learning (AutoML) has been developed with the aim of minimizing manual intervention and automating the creation of ready-to-use machine learning pipelines. Despite the widespread adoption of numerous successful AutoML systems, their performance is observed to vary across different datasets and learning scenarios. Algorithm selection has been proposed as a solution to this limitation, enabling the recommendation of algorithms on a per-instance basis. In this study, algorithm selection is utilized to enhance the capabilities of existing AutoML frameworks with minimal additional effort. A comprehensive empirical evaluation is conducted on 39 diverse tasks from the OpenML platform, involving 6 state-of-the-art AutoML methods. The results demonstrate that the integration of algorithm selection not only amplifies the strengths of current AutoML systems but also leads to improved performance and robustness. These findings highlight the practicality and efficacy of algorithm selection as a critical advancement for the next generation of AutoML technologies.
The resilience of the pharmaceutical supply chain (PSC) is crucial to ensure the continuous availability of essential medicines and other products, particularly during logistics disruptions. This article presents a systematic review of the literature on machine learning (ML) and artificial intelligence (AI) techniques and their importance to PSC resilience. By analyzing existing literature from multiple databases, including Web of Science and IEEE Xplore, over the past five years, we identified key areas where ML and AI have been effectively utilized. These areas include demand forecasting, risk management, and inventory optimization. This review also highlights significant research gaps and proposes future directions for investigation. Our findings suggest that while ML and AI offer promising solutions for improving supply chain resilience, there is a need for more studies that integrate various ML and AI approaches into PSC. Our analysis reveals that there are no clear government regulations related to the usage of ML or AI in PSC, no robust real-world applications addressing the challenges of adopting these technologies, and no clear predictive models to assess their impact on PSC resilience. This ongoing work aims to provide a foundation for future research, ultimately fostering more adaptive and resilient PSCs.
The Knapsack Problem (KP) is a classic combinatorial optimization challenge with applications in several fields including logistics and finance. Traditional solving approaches often struggle to solve large-scale and complex instances, necessitating the adoption of advanced metaheuristic techniques. This study presents a novel approach that integrates chaotic maps into the Tasmanian Devil Optimization (TDO) algorithm to address the KP more effectively. Specifically, a chaotic binarization mechanism is proposed to improve the balance between exploration and exploitation. Comparative analyses with state-of-the-art metaheuristics were performed, using KP benchmark datasets. The results demonstrate that the chaotic sine-based TDO outperforms traditional methods in terms of solution quality, convergence stability, and computational efficiency. These findings highlight the potential of chaotic hybridization as a promising tool for solving binary combinatorial optimization problems.
With the increasing volume of data and diverse user requirements, the cloud-edge-end architecture must adapt flexibly to address these challenges. However, many existing studies overlook the complexities of heterogeneous networks and task dependencies, leading to a mismatch between computational models and real-world applications. This paper addresses these issues by considering heterogeneity, diversity, and dependencies. We first establish a heterogeneous computing environment, and model the complex task dependencies using a directed acyclic graph (DAG). We then propose an improved evolutionary algorithm for task offloading and resource allocation. Experimental results show that our method enhances resource utilization and overall system performance in heterogeneous environments.
The purpose of this paper is to propose a comprehensive methodology for reducing mold changeover time in the plastic injection molding industry by using Lean Manufacturing tools, the approach, and SMED techniques enhanced with artificial intelligence (AI) for decision- making and optimization. First, we will identify the problem using the 5-Why tool. The measurement phase involves the collection of data using time studies and computer-aided design (CAD) to identify failures. In the analysis phase, we will try to address critical tasks and support our analysis with Failure Mode and Effect Analysis (FMEA). The innovate phase enables us to improve the sequence of external and internal tasks, using real-time data to adjust the process. In addition, the methodology helped to balance the production line by redistributing tasks and reducing bottle- necks, hence improving flow and reducing downtime. Application of this approach not only reduced mold changeover time by 65
This paper addresses the two-stage supply chain network design (TS-SCND) problem with fixed costs, that models a distribution network in a supply chain that contains manufacturers, distribution centers (DCs) and retailers. We formulate the investigated problem as a mixed linear integer programming model that contains two kinds of fixed costs: ones corresponding to the distribution routes and the others corresponding to opening the DCs. In this study, our aim is to examine the characteristics of the existing metaheuristic algorithms for solving the investigated problem: four ant colony optimization algorithms and a hybrid genetic algorithm. Based on the performance comparison results, we conduct a systematic analysis of the considered solution approaches. The performed analysis points out that the hybrid genetic algorithm has the best performance regarding the count of obtained optimal solutions and the relative percentage deviations. These findings were validated through statistical analysis.
Recommendation systems (RS) have become essential for providing personalized user experiences across various platforms, suggesting items ranging from products to media content. In recent years, the incorporation of Generative Adversarial Networks (GANs), a groundbreaking generative modeling technique introduced in 2014, has significantly enhanced the capabilities of RS. Despite the promise of GANs in recommendation systems, their performance is highly sensitive to the choice of hyperparameters. In this paper, we aim to optimize the hyperparameters of a GAN-based approach for learning user latent factors in matrix factorization, with the goal of improving top-N recommendations. Our study evaluates different Hyperparameters Optimization methods, additionally, we will study the advantages and limitations of each method.
This paper explores the use of Optimisation Modulo Theories (OMT) to address classic Constraint Satisfaction (and Optimisation) Problems. By leveraging the Z3 solver, we investigate the impact of different SMT logics—Linear Integer Arithmetic (LIA), Linear Real Arithmetic (LRA), Quantifier-Free Bit-Vectors (QF_BV), and Quantifier-Free Arrays with Integer Arithmetic (QF_ALIA)—on problem-solving efficiency and solution quality. A set of automatic encoding rules is proposed to transform problems initially modelled in LIA into other logics, simplifying the process for users without specialised expertise. We demonstrate how logic selection can significantly influence solver performance. For example, QF_BV excels in scheduling and sequential problems like the Nurse Scheduling Problem and the Travelling Salesman Problem. In contrast, LRA and QF_ALIA stand out in problems with broad numerical ranges, such as the Unbounded Knapsack Problem. These findings underscore the importance of an automatic logic encoder capable of dynamically selecting and transforming models into the most appropriate SMT logic for a given problem. This performance-driven logic selection approach enhances the solving process. It extends the accessibility and practicality of OMT frameworks across diverse problem domains.
This study explores the integration of the Kolmogorov-Arnold Network (KAN) model into continuous reinforcement learning (CRL). We incorporate KAN and its improved variant, MultKAN, into the on-policy PPO algorithm, resulting in a new approach called KAN-PPO. Using the Mujoco Half-cheetah environment, we conduct extensive experiments to compare the performance and generalization of KAN-PPO models against MLP-PPO structures. Additionally, we examine the effects of dropout regularization through the DropKAN method in dense network settings. Our findings demonstrate the effectiveness of KAN and MultKAN models in CRL, highlighting their potential for complex learning tasks.
The transportation of hazardous materials (hazmat) poses significant risks to urban areas, necessitating careful planning to minimize potential exposure. This study addresses a variant of the Bilevel Hazmat Network Design Problem, introducing an additional objective to balance the usage of certain roads, thereby preventing the overexposure of specific roads and zones within the city. The problem is mathematically formulated as a semivectorial bilevel optimization model, where the leader’s objective seeks to minimize risk, and the follower optimizes two objectives: minimizing transportation costs and balancing road usage. To address the challenge of determining the follower’s optimal reaction to the leader’s decisions, we obtain a Pareto front for the follower’s objectives. To simplify this process, we make a realistic assumption regarding the follower’s response and propose a heuristic deconstructive algorithm tailored to the problem’s complex structure. Computational experiments are conducted on instances derived from real-life case study to evaluate the performance of the proposed approach. The results demonstrate the heuristic’s effectiveness in designing low-risk networks based on the hazmat flow determined by the follower. Additionally, managerial insights are provided, emphasizing the approach’s practical applicability to urban hazmat management.
Large Language Models (LLMs) have revolutionized natural language processing (NLP) by achieving state-of-the-art performance across diverse tasks. However, fine-tuning these models for domain-specific applications is significantly constrained by the computational costs associated with their training. In this paper, we propose two complementary approaches to address the Hyperparameter Optimization (HPO) challenge in LLM fine-tuning: Bayesian Optimization based on Gaussian Process (BO-GP) and Partition-Based Optimization (PBO). On the one hand, BO efficiently exploits historical knowledge to achieve optimal results within a limited number of evaluations, but its inherently sequential nature poses scalability challenges. On the other hand, PBO enables massive parallelization, making it more scalable but requiring significantly more evaluations to converge. To leverage their complementary strengths for optimizing expensive objective functions, we investigate these methods and propose a hybrid BO-PBO algorithm. This work represents a foundational step toward harnessing the potential of parallel Bayesian Optimization-based algorithms for solving expensive optimization problems in exascale computing environments.
Promotion planning for a store group is a strategic approach aimed at temporarily lowering prices to stimulate sales over a specific period while meeting revenue and margin targets. The challenge lies in selecting the optimal set of products and bundles for promotion and determining their discounts, all within constraints like budget limits, bundle conflicts, and business rules. Conventional methods using linearization or heuristics often lead to sub-optimal solutions, making them unsuitable for real-world scenarios. This paper introduces a two-step Mixed-Integer Nonlinear Programming (MINLP) solution to address the non-linearities inherent in promotion planning. By integrating discrete decisions, such as item selection, with continuous variables like discount levels, MINLP solution offers a comprehensive solution for balancing revenue and margin trade-offs. It allows inclusion of important factors such as hoarding, new bundles and cannibalization effects. In practical settings, where promotions involve about 50–100 SKUs and bundles within a category, the resulting MINLP problem remains computationally feasible and scalable. Compared to the historical discount from a large retailer, promotion using MINLP gives 0.84
This paper introduces a preliminary agent-based model designed to simulate trust dynamics within a multi-agent scenario where virtual agents navigate unknown environments. The objective is to investigate how trust influences agent performance when they rely on an external source of information, referred to as a robot, versus relying on their peers. In the model, the robot explores the environment and reports the costs associated with different paths. Agents are categorized as either trustful or skeptical, deciding whether to trust the information provided by the robot or rely on their fellow agents. We evaluate three distinct scenarios: No Robot (baseline exploration), Immediate Start (simultaneous exploration by the robot and agents), and Early Start (robot-initiated exploration preceding agents’ involvement). Evaluation metrics include the number of agents successfully reaching the exit, exit times, and path costs. Our findings demonstrate significant advantages associated with the presence of the robot, particularly when exploration begins early. Trustful agents show improved performance by optimizing both the number of successful exits and the efficiency of path selection, resulting in reduced exit times and path costs.
Supplier selection is a fundamental aspect of supply chain management (SCM), playing a crucial role in enhancing operational efficiency, cost management, and overall performance. This study evaluates supplier performance using advanced data analytics and predictive modeling techniques. Leveraging a comprehensive dataset sourced from Microsoft Ignite, key performance metrics such as defect quantity, downtime, and categorical identifiers were analyzed to uncover trends and actionable insights. Visualizations created using Power BI dashboards highlighted critical performance trends, guiding targeted interventions for operational improvement. Various methodologies, including logistic regression, multinomial and ordinal regression, random forest models, and Gaussian Naive Bayes, were applied using Minitab, Power BI, and Python to explore the relationships between predictor variables and performance outcomes. The initial analysis revealed limitations in linear models due to weak correlations among categorical variables, prompting the adoption of multinomial and ordinal logistic approaches for improved classification accuracy. This study demonstrates how robust modeling and visualization frameworks can inform strategic decision-making in supply chain management, fostering sustainable and adaptive supplier relationships. Despite these advancements, the findings emphasize the necessity for further refinement to address complex interdependencies and enhance predictive accuracy.
This paper deals with the gantry cranes problem scheduling at a container terminal. The scheduling of this equipment aims to minimize container assignment time for two different operations. A mixed integer linear programming model is developed to minimize execution time. An optimized greedy heuristic algorithm (GHA) is also applied to solve this problem, with a new best-scheduling update strategy included to improve solution stability. Results show that the proposed GHA efficiently solves the RTG cranes scheduling problem.
Feature selection allows reducing the dimensionality of data sets by selecting the most relevant features. This will lead to more interpretable models, in addition to reducing computational costs. In this context, we introduce a new nature-inspired feature selection technique based on a recently developed discrete optimization algorithm called the Binary Mother Tree Optimization (BMTO) algorithm. We call BMTO for Feature Selection (BMTO-FS) the new technique we propose. BMTO-FS utilizes a binary converter module that converts real numbers into binary numbers using a sigmoid function and a threshold value. BMO-FS operates within our new feature selection framework that can be generalized to any other metaheuristic. To assess the performance of BMTO-FS, we conducted extensive comparative experiments on several public benchmark datasets corresponding to low, medium, and high dimensions. In these experiments, genetic algorithms (GA), particle swarm optimization (PSO), and the whale optimization algorithm (WOA) are considered. The results are promising, as they show that BMTO-FS outperforms GAs, PSO, and WOA in terms of accuracy rates, number of reduced features, and computational cost.
Effective representation and understanding of user preferences play a vital role in various applications such as recommender systems, product customization, resource allocation, configuration, and online auctions. Quantitative preferences are often represented using utility functions, while formalisms such as the semi-ring-based Constraint Satisfaction Problem (CSP) and the Valued CSP extend the CSP framework to handle both quantitative preferences and constraints. However, it might not always be convenient for users to quantify their preferences. A qualitative preference ordering can be more natural in these circumstances. The Conditional Preference Network (CP-net) graphical model is designed to express users’ conditional preferences through Conditional Preference Tables (CPTs). In many real-world combinatorial applications, conditional preferences coexist with hard constraints. In this context, the CP-net has been extended to constraints through the constrained CP-net (CCP-net) model. Solving the CCP-net consists in finding a set of Pareto-optimal solutions that meet all constraints while optimizing the qualitative preferences. Finding Pareto solutions in CCP-nets using exact methods like branch and bound can be time-consuming. Alternative methods, trading the quality of the returned solutions for the execution time, can be found in metaheuristics, including nature-inspired techniques. We introduce three bio-inspired methods to enhance the efficiency of finding Pareto solutions in CCP-nets. Additionally, we experimentally compare these techniques with those derived from an exact method, focusing on the number of Pareto solutions identified and the runtime performance. The results of the experiments, conducted on random CCP-net instances generated by the RB model, are reported and discussed.
