Many optimization algorithms struggle to escape local optima in high-dimensional problems. To overcome this problem, this paper introduces a single objective swarm-based evolutionary metaheuristic algorithm inspired by behavior of hooded seals, called Hooded Seal optimization (HoSeO). The proposed algorithm demonstrates a favorable balance between exploration and exploitation, thereby addressing the aforementioned problem. The underlying mathematical structure of HoSeO consists of three main procedures: a mating phase based on proximity-strength pairing, mentoring phase levy flight with adaptive perturbations, along with a stochastic seasonal movement phase, each guided by dynamic parameter control. A unique population update strategy was utilized that combines elite, offspring, and remaining top candidates to preserve both solution quality and diversity. To evaluate performance, HoSeO was tested on 23 classical benchmarks, 10 CEC-2019 functions, 12 CEC-2022 functions, and six real-world problems. The achieved results were then compared to well-known algorithms such as Genetic Algorithm (GA), Multi-Verse Optimizer (MVO), White Shark Optimizer (WSO), Learner Performance-based Behavior (LPB), Dragonfly Algorithm (DA), Linear Population Size Reduction Success-History Based Parameter Adaptation Differential Evolution (L-SHADE), and Raindrop Algorithm (RD). Experimental results demonstrate that HoSeO consistently delivers superior performance, outperforming or matching competitive algorithms on 17 out of 23 classical benchmark functions and achieving dominance in 7 out of 10 CEC 2019 test suite. Additionally, HoSeO ranked among the top-performing methods in 9 out of 12 CEC 2022 benchmarks. To further validate its robustness and practical applicability, the algorithm was tested on six real-world problems—highlighting its strong capability to handle complex and diverse optimization tasks effectively. Furthermore, the importance of the results was confirmed through statistical validations.
Imaging techniques such as chest X-rays, whole slide images (WSI), and optical coherence tomography (OCT) serve as primary screening and detection tools for a wide variety of pulmonary and ophthalmic conditions. This paper presents a systematic and unified empirical investigation of the impact of ImageNet-pretrained deep convolutional neural networks on medical image classification across three distinct imaging modalities. The study evaluates ten widely adopted CNN architectures under five training strategies: three transfer learning configurations for binary and multiclass classification based on ImageNet pretraining (frozen feature extraction, partial fine-tuning, and full fine-tuning) and two randomly initialized configurations (partial training and full training from scratch). The proposed framework was evaluated on three publicly available medical imaging datasets. For binary classification, the chest X-ray pneumonia dataset consisted of 4,000 training, 1,043 validation, and 624 testing images. For multiclass classification, experiments were performed on the OCT retinal dataset with 86,647 training, 21,662 validation, and 1,000 testing images, and the LC25000 histopathology WSI dataset with 9,600 training, 1,680 validation, and 720 testing images. The model performance was evaluated using AUC, precision, sensitivity, and specificity. All metrics were reported as the mean ± SD along with their corresponding 95
Facial Expression Recognition (FER) is an important research area in computer vision and affective computing, with challenges in recognizing micro-expressions and subtle expressions across cultures. Existing models often face challenges in generalization because they are trained on facial expression datasets drawn from restricted geographic regions, and as a result, struggle to interpret expressive patterns that may be influenced by cultural differences. Indian aesthetics defines Navarasa as nine subtle emotional states that change slowly and are hard to capture. This paper presents a generalized, compact, and lightweight four-layer Convolutional Neural Network (CNN) architecture that incorporates Efficient Channel Attention (ECA) and Squeeze-and-Excitation (SE) mechanisms for real-time Navarasa classification, referred to as NavLite-AttNet. ECA is applied in layers 1-3 to enhance localized channel interactions in shallow convolution blocks, while SE in layers 3-4 captures global affective dependencies. The hybrid use of ECA followed by SE in layer 3 further refines intermediate features before deeper processing. During preprocessing, a Multi-Task Cascaded Convolutional Network (MTCNN) was used for accurate face detection. NavLite-AttNet outperforms the baseline four-layer CNN by +5.16
The necessity for rapid and accurate estimation of machining indicators for sustainable wire electrical discharge machining (WEDM) makes modeling these indicators and simultaneous parameter optimization indispensable. This study aims to develop artificial neural network (ANN) models and optimize parameters to minimize surface roughness (Ra), kerf width (Kw), and total energy consumption (Pc) during WEDM of hardened D2 steel. WEDM experiments were conducted using the Taguchi orthogonal array generated from the parameters of pulse width (Ton), pulse space (Toff), wire speed (Wf), and feed rate (f). In the modeling process, the experimental data were separated into training, test, and validation datasets. Feed-forward back propagation network and Levenberg–Marquardt algorithm were applied as training functions to predict the indicators. The most striking findings are that increases in Ton and Wf increase kerf width and Ra, while increases in feed rate significantly reduce Kw and energy consumption. A 52.27
The proliferation of deep learning applications in several areas has led to the rapid adoption of such solutions from an ever-growing number of institutions and companies. These entities’ Deep Neural Network (DNN) models are often trained on proprietary data. They require powerful computational resources, with the resulting DNN models incorporated into the company’s work pipeline or provided as a service. Being trained on proprietary information, these models offer a competitive edge for the owner company. At the same time, these models can be attractive to competitors (or malicious entities), which can employ state-of-the-art security attacks to obtain and use these models for their benefit. As these attacks are hard to prevent, it becomes imperative to have mechanisms that enable an affected entity to verify the ownership of its DNN with high confidence. This paper presents TATTOOED, a robust and efficient DNN watermarking technique based on spread-spectrum channel coding. TATTOOED has a negligible effect on the performance of the DNN model. We extensively evaluate TATTOOED against several state-of-the-art mechanisms used to remove watermarks from DNNs and found it robust to such removal techniques even in extreme scenarios. For example, even if removal techniques alter as much as 99
