The stochastic contextual bandit problem, recognized for its effectiveness in navigating the classic exploration-exploitation dilemma through ongoing player-environment interactions, has found broad applications across various industries. This utility largely stems from the algorithms’ ability to accurately forecast reward functions and maintain an optimal balance between exploration and exploitation, contingent upon the precise selection and calibration of hyperparameters. However, the inherently dynamic and real-time nature of bandit environments significantly complicates hyperparameter tuning, rendering traditional offline methods inadequate. While specialized methods have been developed to overcome these challenges, they often face three primary issues: difficulty in adaptively learning hyperparameters in ever-changing environments, inability to simultaneously optimize multiple hyperparameters for complex models, and inefficiencies in data utilization and knowledge transfer from analogous tasks. To tackle these hurdles, this paper introduces an innovative transfer learning-based approach designed to harness past task knowledge for accelerated optimization and dynamically optimize multiple hyperparameters, making it well-suited for fluctuating environments. The method employs a dual Gaussian meta-model strategy—one for transfer learning and the other for assessing hyperparameters’ performance within the current task —enabling it to leverage insights from previous tasks while quickly adapting to new environmental changes. Furthermore, the framework’s meta-model-centric architecture enables simultaneous optimization of multiple hyperparameters. Experimental evaluations demonstrate that this approach markedly outperforms competing methods in scenarios with perturbations and exhibits superior performance in 70% of stationary cases while matching performance in the remaining 30%. This superiority in performance, coupled with its computational efficiency on par with existing alternatives, positions it as a superior and practical solution for optimizing hyperparameters in contextual bandit settings.

In recent years, the adoption of advanced machine learning techniques has revolutionized approaches to solving complex problems, such as identifying occurrences of forest fires. Among these techniques, the use of Convolutional Neural Networks (CNNs) combined with ensemble methods is particularly promising. To investigate the feasibility of detecting fires using video streams from Unmanned Aerial Vehicles (UAVs), the lightweight CNN architecture MobileNetV2 was utilized for real-time detection. Several experiments were conducted on the DeepFire dataset, which comprises an equal number of images with and without fire, to evaluate MobileNetV2's performance. Notably, the architecture's linear bottlenecks and the efficient use of inverted residuals ensure high accuracy without compromising on feature extraction capabilities. For a comprehensive assessment, MobileNetV2 was benchmarked against other models, including DenseNet121, EfficientNetV2S, and VGG16. Accuracy was enhanced by averaging predictions through methods such as voting or summing results. As documented in the literature, MobileNetV2 consistently outperforms other architectures in computational efficiency and provides an excellent balance between efficiency and the quality of learned features over multiple epochs. This study underscores the suitability of MobileNetV2 for real-time applications on drones, particularly for the detection of forest fires in resource-constrained environments. The results show that MobileNetV2 achieves the highest accuracy (0.994), sensitivity (0.994), and specificity (0.998) among the tested models, with low standard deviations across all metrics. In contrast, EfficientNetV2S exhibited the lowest accuracy and sensitivity, both at 0.779, with a specificity of 0.829. The ensemble (Sum) method achieved an average accuracy of 0.989, sensitivity of 0.989, and specificity of approximately 0.988. Therefore, MobileNetV2 not only delivers the highest accuracy and stability but also demonstrates that the choice of ensemble method significantly affects the results.

Dental caries is arguably the most persistent dental condition that affects most people over their lives. Carious lesions are commonly diagnosed by dentists using clinical and visual examination along with oral radiographs. In many circumstances, dental caries is challenging to detect with photography and might be mistaken as shadows for various reasons, including poor photo quality. However, with the introduction of Artificial Intelligence and robotic systems in dentistry, photographs can be a helpful tool in oral epidemiological research for the assessment of dental caries prevalence among the population. It can be used particularly to create a new automated approach to calculate DMF (Decay, Missing, Filled) index score. In this paper, an autonomous diagnostic approach for detecting dental cavities in photos is developed using deep learning algorithms and ensemble methods. The proposed technique employs a set of pretrained models including Xception, VGG16, VGG19, and DenseNet121 to extract essential characteristics from photographs and to classify images as either normal or caries. Then, two ensemble learning methods, E- majority and E-sum, are employed based on majority voting and sum rule to boost the performances of the individual pretrained model. Experiments are conducted on 50 images with data augmentation for normal and caries images, the employed E-majority and E-sum achieved an accuracy score of 96% and 97%, respectively. The obtained results demonstrate the superiority of the proposed ensemble framework in the detection of caries. Furthermore, this framework is a step toward constructing a fully automated, efficient decision support system to be used in the dentistry area.