The increasing energy demand necessitates efficient renewable energy solutions. This necessity highlights solar energy as the most reliable and abundant source. Since material design is very important for obtaining efficient energy from the sun, metamaterial absorbers are of interest. Metamaterials, which are characterized by their negative refractive index and electromagnetic manipulation capabilities, are preferred in solar absorber designs due to their high efficiency, low cost, and ease of fabrication. The characteristic properties of metamaterial absorbers can be obtained by optimizing the parameters in their geometric structures. However, optimizing these structures requires electromagnetic simulations involving a large parameter space. This situation results in high computational costs that require high-performance computing (HPC) resources and parallel processing. To overcome this challenge, computational steps can be significantly reduced by integrating artificial intelligence tools into optimization processes. In this study, a metamaterial absorber operating at ultraviolet, visible light, and infrared wavelengths is designed, and its solar energy collection efficiency is investigated. Then, a deep learning-based optimization approach is presented for the proposed solar energy absorber. The approach proposed in the study is particularly suitable for HPC-supported environments where large-scale simulations can be generated and processed using parallel computing. With the proposed deep neural network optimization model, geometric parameters are suggested for the maximum absorption value to be obtained from the solar absorber. Thanks to this state-of-the-art deep learning-based optimization model, the number of simulations has been reduced by approximately 90%, decreasing from 1256 to 126. Therefore, the approach proposed in the study significantly reduces the computational workload and reveals the potential of combining deep learning methods with HPC infrastructures in accelerating electromagnetic optimization problems.

Multiple sclerosis (MS) is an autoimmune, neurodegenerative, chronic disease that affects the central nervous system and manifests itself with attacks. Although there is no definite cure for the disease, it is possible to control these attacks. Follow-up of the disease has great importance in terms of disability. An Extended Disability Status Scale (EDSS) is used to show how much the disease affects. This score is determined by specialized clinicians. In this study, the EDSS score, previously determined by neurologists, was attempted to be estimated using the EEG signals. 32-channel EEG signals were recorded while 17 MS patients with EDSS 1.0, 1.5, and 2.0 were performing a working memory task. Using the band power of these 6-minute EEG signals, EDSS estimation was performed with the Decision Tree Regressor, resulting in a Mean Absolute Error (MAE) of 0.088. With the Leave One Out Cross-Validation, 17 trees were extracted and 12 were found to be identical. As a result, the band power features of F7 and CP2 EEG channels were found to be successful in predicting 3-level EDSS scores with a decision tree regressor with 0.0 MAE. Additionally, the relationship between the scores obtained in the working memory task and the EDSS scores of MS patients was statistically calculated with One-way ANOVA. There was no significant difference between the EDSS score and the task scores (p>.05).

Epilepsy is a neurological disorder characterized by recurrent seizures, and EEG continues to be the gold standard for diagnosis in clinical practice. However, the nonlinear nature of the EEG and its sensitivity to noise make manual interpretations difficult. This study presents a feature selection and hyperparameter optimization approach for automatic multi-class EEG-based epilepsy detection. Five classes representing two healthy states and three epileptic states were analyzed using EEG signals from the Bonn database. Temporal, statistical, fractal, and spectral features were extracted, and their importance was ranked according to ANOVA F-values. Classifications were performed using support vector machines, random forests, nearest-neighbor algorithms, decision trees, multilayer perceptrons, and linear discriminant analysis; hyperparameters were optimized using Grid Search Cross-Validation. The robustness of the proposed model was ensured using grouped layered k-fold cross-validation. With this approach, it achieved an F1-score of 0.8126 in 5-class classification and 0.9894 in binary classification, outperforming existing methods on the same dataset. Hjorth complexity, standard deviation, and Hjorth mobility emerged as the most discriminative features for epileptic states. The study demonstrates that combining feature selection with hyperparameter optimization provides a robust, interpretable framework for multi-class EEG classification. This approach has the potential to support neurologists in their epilepsy assessments by improving diagnostic accuracy and providing clinically meaningful insights.