
In this work, we investigate an approach that enriches the information encoded within the ACO pheromone structure in order to enhance the performance of swarm optimization. While this concept has been extensively studied in the context of single-objective optimization, this paper provides a comprehensive evaluation of its extension to multi-objective optimization problems. The results demonstrate that this enhanced pheromone representation remains effective in multi-objective settings.
Dimensionality reduction is a key stage in data processing. Eliminating irrelevant attributes from the input dataset enables more efficient processing and supports the development of robust Machine Learning models. In particular, when dealing with large datasets, attribute reduction algorithms can significantly reduce computational complexity while minimizing the loss of valuable information contained in the data. This paper explores the potential application of a Feature selection method based on Rough Set Theory for multivariate data. The experiments were conducted on the publicly available Taiwan Credit dataset. The results show that limiting the input feature space to a minimal set of attributes generated by a reduct selection algorithm not only reduces training and inference time for Machine Learning or Deep Learning models but may also improve their classification performance. The models achieved higher Accuracy and better adaptability to the input data, as confirmed by increased or comparable Precision and Recall scores. Additionally, the findings indicate that applying Feature selection using a reduct set can lead to more efficient data processing. These results suggest that integrating Rough Set Theory with Deep Learning models still represents a promising approach to analyzing attribute-value data.
Accurate segmentation of brain tumors in magnetic resonance imaging (MRI) is essential for clinical diagnosis and treatment planning. However, variations in tumor morphology, location, and modality-specific appearance make the task highly complex for conventional deep learning approaches. This study introduces SOFAN-UNet, a novel segmentation framework that improves tumor localization and boundary delineation through hybrid attention mechanisms and bio-inspired optimization. The proposed SOFAN-UNet architecture builds on the traditional U-Net structure and incorporates a hybrid attention mechanism combining Squeeze-and-Excitation (SE) and Self-Attention modules. A short optimization strategy, implemented via Particle Swarm Optimization (PSO), is used to dynamically calibrate attention gate parameters, including scale factors and filter sizes, within a constrained search space. This enables efficient learning of discriminative tumor features while preserving spatial detail through optimized skip connections. The model was evaluated across three benchmark datasets-FBTS, BraTS 2021, and BraTS 2018-using DSC, JI, HD, ASSD, AUC, and MCC metrics. Ablation studies and five-fold cross-validation were conducted to assess robustness. SOFAN-UNet achieved state-of-the-art performance, with Dice scores reaching 0.9685 and Jaccard Index up to 0.9389. It demonstrated strong generalization across MRI modalities and tumor types, maintained low boundary error (HD < 2.5), and achieved AUC values above 0.999. Ablation results confirmed the effectiveness of PSO-based calibration and attention integration, especially for heterogeneous tumor regions. The SOFAN-UNet model offers a highly accurate and computationally efficient framework for brain tumor segmentation. Its robustness across datasets and imaging modalities supports its potential for clinical deployment and further extension to subregion tumor analysis.
The paper presents the application of the Bees Algorithm to solve the NP -hard optimisation problem related to the Belt Conveyor Systems. For this purpose, the mathematical model for a real-life example of the Belt Conveyor Production System (BCPS) was defined. Creating the multi-swarm Bees Algorithm for solving this optimisation problem on a CUDA platform within the accepted time for real production systems, and analyzing the impact of control parameters on the algorithm's performance are the main objectives of our study. Moreover, the impact of the different block and grid sizes was checked. To ensure comparability of the results, extensive tests have been carried out using the multi-swarm version designed for the GPU and the mono-swarm version designed for the CPU platform. The results show that the proposed multi-swarm Bees Algorithm algorithm for the CUDA platform gives considerable acceleration of the calculations, but often requires the larger size of the neighbourhood.
Multi-agent dynamic task allocation (MADTA) for UAV swarm and autonomous systems remains a formidable challenge in highly uncertain and stochastic environments, where conventional reinforcement learning methods struggle with variable input dimensions and coordination conflicts. This paper proposes a spatiotemporal topology-aware graph reinforcement learning (STA-GRL) framework to address these limitations. By modeling the environment as a dynamic bipartite graph, the framework integrates a spatiotemporal gated graph attention (STGGA) module that employs a temporal gating mechanism to dynamically prioritize tasks with rapidly decaying deadlines. A topology-aware critic is further designed to penalize spatial conflicts among agents via an enhanced adjacency matrix. Extensive simulations demonstrate that STA-GRL significantly surpasses state-of-the-art baselines. In the primary evaluation scenario with 30 agents and an intermediate task arrival rate (lambda = 0.6), STA-GRL achieves a task completion rate of 86.8% and an average response time of 18.4 seconds, while reducing the average conflict rate to just 2.1%. Moreover, ablation studies confirm the critical contribution of each architectural component, with the temporal gate improving the completion rate by 7.3% and the topology-aware critic reducing conflicts by 6.4%.