Third molar segmentation on panoramic dental radiographs is crucial for oral diagnosis and effective treatment planning. However, it remains a challenging task due to anatomical complexity, variability, and frequent image artifacts. This work presents a comprehensive benchmarking of state-ofthe-art semantic segmentation models from three architecture families: Convolutional Neural Networks (CNNs), Transformer-based models, and Hybrid architectures. We introduce a novel dataset of 871 annotated panoramic radiographs, labeled with four third molar classes based on the FDI numbering system. The study evaluates U-Net (with MobileNetV2, ResNet34, ResNet50), DeepLabV3, SegFormer, SETR, DETR, and Mask2Former models. All models were trained and compared under a consistent experimental setup with extensive hyperparameter tuning. Performance was assessed using multiple metrics: Mean IoU, Dice Score, Precision, Recall, and Pixel Accuracy. U-Net with ResNet34 achieved the highest segmentation performance (Mean IoU: 0.7771, Dice: 0.8691), closely followed by SegFormer. These results highlight the superior spatial localization of CNNs and the contextual modeling strength of Transformers. In contrast, hybrid models such as DETR and Mask2Former ranked lower, indicating limitations in capturing smaller and irregular anatomical structures in dental radiographs. The outcome of this study offer important insight into the challenges of applying various deep learning paradigms to dental image analysis and underscore the significance of architectural choices and dataset properties in achieving clinically relevant outcomes.

Semantic segmentation in complex natural environments remains a challenging task due to diverse object appearances, background clutter, and class imbalance. This study evaluates state-of-the-art segmentation models, categorized into UNet-based, encoder-decoder, and transformer-based architectures, using a dataset focused on aquatic plant segmentation in a natural environment. The model is evaluated using Dice coefficient, F1 score, IoU and overall accuracy, along with computational complexity measured by parameter count and GFLOPs. Results indicate that backbone-enhanced models outperform base configurations by providing deeper and more robust feature representations. Then, explainable AI (XAI) techniques are used to evaluate the model attention and ensure the precise focus of the network on ecologically meaningful regions. These insights not only highlight the importance of architectural depth and input diversity but also demonstrate how XAI can support iterative model refinement. As a result of these combined efforts, segmentation performance is improved from a baseline dice coefficient and accuracy of 50.76 and 85.07 % to a maximum 75.57 and 94.94 % respectively. This comprehensive analysis provides valuable insights of selecting suitable models in resource-constrained environments and advances the development of robust segmentation systems.

Stress among university students can be defined as a state of mental tension caused by difficult academic endeavours. Nonetheless, stress at the tertiary (i.e., University) level has been the subject of much research as accurately predicting mental stress is crucial for promoting student well-being and academic success. In recent research, Machine Learning (ML) and Deep Learning (DL) models have been utilized for determining the academic stress. However, majority of these works utilizes isolated ML or DL models without having an in-depth assessment of the feature interpretation (i.e., the mental stressors) with respect to model performance. In this study, a novel hybrid CNN-SVM model is proposed to predict the mental stress among the university students. The rationale here is that the feature extraction capability of Convolutional Neural Networks (CNN) blended with the classification proficiency of Support Vector Machines (SVM) would better extract the relevant mental stressors towards accurate stress prediction. The model is trained and tested with the student mental stress data as measured using the PSS-10 model from 15 top-ranked universities in Bangladesh. Alongside, Explainable Artificial Intelligence (XAI) is used to validate the model performance. Reported results reveal superior performance of the hybrid CNN-SVM model in relation to conventional ML and DL models, achieving an accuracy of 97.34% with precision 97.38%, recall 97.34%, and f1-score 97.36%. XAI assessment shows that the improvement is due to the models' efficacy in utilizing the essential mental stressors than the conventional models. We argue that the overall contribution can be used for early detection of student mental stress and assist in planning for student counseling, health assessment, and academic process improvement.

This paper addresses, in the context of wireless network design, the problem of optimally partitioning the spectrum into a set of nonoverlapping channels with nonuniform spectrum widths. While narrower bands split the total available spectrum into more nonoverlapping channels allowing more parallel concurrent transmissions, wider spectrum bands yield links with larger transport capacity. Thus, we model the combinatorially complex problem of joint routing, link scheduling, and variable-width channel allocation in both single and multirate multihop wireless networks as a mixed integer linear program, and present a solution framework using the column generation decomposition approach. Given the nature and complexity of the resulting dual subproblem, we propose heuristic methods for partitioning the spectrum and allocating resources to each active links, and hence obtain solutions for larger network instances. We present several numerical results and engineering insights suggesting both spectrum width and transmission rates as effective tunable knobs for combatting interference and promoting spatial reuse and thus achieving superior performance in multihop settings.

We investigate the joint problem of relay selection and optimal sharing of relay power in wireless cellular networks with multicast traffic. We use two different performance metrics to maximize the network performance. We first present a mixed Boolean-convex optimization model to maximize the overall network capacity and solve this combinatorial problem optimally using branch and bound technique. We then show that obtaining the optimal solution is computationally not feasible for large network sizes and, unlike the case of unicast traffic, a water filling method does not yield near optimal solutions in multicast scenarios. We thus adopt an algorithm based on sequential fixing which substantially reduces the computation time and achieves near optimal solutions. In addition, we present a mixed integer linear programming model to maximize the capacity of the minimum capacity link and show that the model is very efficient to reach the optimal solutions.

We investigate the problem of flexible spectrum access in multihop wireless networks. We assume radios that are capable of transmitting on channels of contiguous frequency bands and which do not require any sophisticated processing. Because these radios can flexibly configure their transmissions anywhere in the available frequency band, the spectrum becomes vulnerable to fragmentation and interference. We consider the joint problem of routing, link scheduling and spectrum allocation where scheduling feasibility is considered under the physical interference (SINR) constraint. We present a primal-dual decomposition for this complex optimization problem based on column generation. We show that obtaining the optimal solution to this problem is computationally not feasible, except for very small networks. We thus adopt a two-fold method to circumvent the complexity while yielding practical solutions. First, we relax the SINR constraint and use a simplified graph-based model for the interference. Second, we use a simulated annealing (SA) approach to solve the dual subproblem. Our SA approach however is augmented with an SINR feasibility check. Our results confirm that the primal-dual decomposition method using SA substantially reduces the computation time and achieves near optimal solutions. The results also reveal that substantial improvement in network performance is obtained with flexible spectrum assignment which results from its capability of better managing the interference in the network.

Recent research results have shown that channel width is an important control knob that can be easily adapted through software and can be used for achieving higher system throughput and better energy efficiency. In this paper we address the problem of joint routing and transmission scheduling in a multichannel wireless mesh network with variable-width channel allocation. While narrower bands split the total available spectrum into more non-overlapping channels allowing more parallel concurrent transmissions, wider bands increase the capacity of the communication links. We present a cross-layer problem formulation which incorporates multi-path routing and link layer scheduling. We model this combinatorial complex problem as a mixed integer linear program and present a primal-dual decomposition method for solving it. Our method has always shown to strike a good balance between often conflicting objectives to achieve higher system performance. Numerical results revealed that up to 57% improvement in network performance is obtained when variable-width channel assignment is used against the best fixed-width channel assignment for larger networks; this is due to the capability of the former in achieving a good balance between higher concurrency and better control of interference.

We investigate the problem of scheduling and flexible spectrum access in multi-hop networks with software defined radios. We present a primal-dual decomposition to provide an exact solution for this complex optimization problem. We show that obtaining the optimal solution is computationally not feasible. Thus, we first relax the SINR constraint and use a simplified graph-based interference model. Second, we use a simulated annealing (SA) approach for solving the dual sub-problem. Our SA approach is then augmented with a feasibility check so that only SINR-feasible schedules are passed back to the primal.

We investigate the use of a binary version of ant colony optimization for the cryptanalysis of an identification scheme based on the permuted perceptron problem (PPP) proposed by Pointcheval. Based on our experimental results, ACO-based attacks proved to be very effective on recovering the secret key of this scheme for various sets of PPP parameters.

We investigate the use of Particle Swarm Optimization (PSO) in automated cryptanalysis of classical simple substitution ciphers. Based on our experimental results, PSO-based attacks proved to be very effective on various sets of encoding keys.

Using a heuristic search technique, several examples for 9-variable Boolean functions with nonlinearity 240, algebraic degree 5, and resiliency degree 3 were constructed. This construction affirmatively answers the open problem about the existence of such functions.