Skin cancer, particularly melanoma, has been recognized as one of the most lethal forms of cancer. Detecting and diagnosing skin lesions accurately can be challenging due to the striking similarities between the various types of skin lesions, such as melanoma and nevi, especially when examining the color images of the skin. However, early diagnosis plays a crucial role in saving lives and reducing the burden on medical resources. Consequently, the development of a robust autonomous system for skin cancer classification becomes imperative. Convolutional neural networks (CNNs) have been widely employed over the past decade to automate cancer diagnosis. Nonetheless, the emergence of the Vision Transformer (ViT) has recently gained a considerable level of popularity in the field and has emerged as a competitive alternative to CNNs. In light of this, the present study proposed an alternative method based on the off-the-shelf ViT for identifying various skin cancer diseases. To evaluate its performance, the proposed method was compared with 11 CNN-based transfer learning methods that have been known to outperform other deep learning techniques that are currently in use. Furthermore, this study addresses the issue of class imbalance within the dataset, a common challenge in skin cancer classification. In addressing this concern, the proposed study leverages the vision transformer and the CNN-based transfer learning models to classify seven distinct types of skin cancers. Through our investigation, we have found that the employment of pre-trained vision transformers achieved an impressive accuracy of 92.14%, surpassing CNN-based transfer learning models across several evaluation metrics for skin cancer diagnosis.

Convolutional encoder-decoder (CED) has emerged as a powerful architecture, particularly in speech enhancement (SE), which aims to improve the intelligibility and quality and intelligibility of noise-contaminated speech. This architecture leverages the strength of the convolutional neural networks (CNNs) in capturing high-level features. Usually, the CED architectures use the gated recurrent unit (GRU) or long-short-term memory (LSTM) as a bottleneck to capture temporal dependencies, enabling a SE model to effectively learn the dynamics and long-term temporal dependencies in the speech signal. However, Transformers neural networks with self-attention effectively capture long-term temporal dependencies. This study proposes a multi-attention bottleneck (MAB) comprised of a self-attention Transformer powered by a time-frequency attention (TFA) module followed by a channel attention module (CAM) to focus on the important features. The proposed bottleneck (MAB) is integrated into a CED architecture and named MAB-CED. The MAB-CED uses an encoder-decoder structure including a shared encoder and two decoders, where one decoder is dedicated to spectral masking and the other is used for spectral mapping. Convolutional Gated Linear Units (ConvGLU) and Deconvolutional Gated Linear Units (DeconvGLU) are used to construct the encoder-decoder framework. The outputs of two decoders are coupled by applying coherent averaging to synthesize the enhanced speech signal. The proposed speech enhancement is examined using two databases, VoiceBank+DEMAND and LibriSpeech. The results show that the proposed speech enhancement outperforms the benchmarks in terms of intelligibility and quality at various input SNRs. This indicates the performance of the proposed MAB-CED at improving the average PESQ by 0.55 (22.85%) with VoiceBank+DEMAND and by 0.58 (23.79%) with LibriSpeech. The average STOI is improved by 9.63% (VoiceBank+DEMAND) and 9.78% ( LibriSpeech) over the noisy mixtures.

The latest manifestation of “all connected world" is the Internet of Things (IoT), and Internet of Vehicles (IoV) is one of the key examples of IoT these days. In Social IoV (SIoV), each vehicle is treated as a social object where it establishes and manages its own Social Network (SN). Incidentally, most of the SIoV research in the literature is related to proximity-based connectivity and interactions. In this paper, we bring people in the loop by incorporating their SNs. While emphasizing a recommendation scenario, in which vehicles may require recommendations from SNs of their owners (in addition to their own SIoV), we proposed an agent-based model of information sharing (for context-based recommendations) on a hypothetical population of smart vehicles. Some important hypotheses were tested using a realistic simulation setting. The simulation results reveal that a recommendation using weak ties is more valuable than a recommendation using strong ties in pure SIoV. The simulation results also demonstrate that recommendations using the most-connected person in the social network are not more valuable than recommendation using a random person in the social network. The model presented in this paper can be used to design a multi-scale recommendation system, which uses SIoV and a typical SN in combination.

This paper presents a novel privacy-preserving architecture, a fusion of Federated Learning with Personalized Models and Differential Privacy (FLPMDP), for diagnosing arrhythmia from 12-lead electrocardiogram (ECG) signals. The architecture supports collaborative training in decentralized healthcare institutions without exposing sensitive patient information. By employing gated recurrent units (GRUs) for temporal sequence modeling along with feature fusion techniques and local differential privacy enforcement, FLPMDP ensures robust classification performance with data confidentiality. The architecture is evaluated on four experimental setups and demonstrates significant performance gain over centralized and federated baseline models. An empirical experiment on a large ECG dataset of 10,646 recordings indicates that the FLPMDP approach achieves an average accuracy of 93.71%. The FLPMDP approach yields F1-scores of 0.98, 0.93, 0.88, and 0.89 for sinus bradycardia (SB), atrial fibrillation (AFIB), supraventricular tachycardia (GSVT), and sinus rhythm (SR), respectively. Additionally, FLPMDP recorded a specificity up to 0.98, with a Kappa score of 0.8971 and a Matthews Correlation Coefficient of 0.9042, indicating high diagnostic accuracy and model strength. Comparative analysis against state-of-the-art methods—such as CNN, ResNet, and attention-based RNNs—indicate that FLPMDP consistently outperforms current models in accuracy, sensitivity, and robustness when facing non-IID data conditions. In the context of this research, federated learning is highly pertinent to modern healthcare, enabling secure and collaborative model training across institutions while complying with data privacy. The proposed FLPMDP framework offers a scalable and privacy-compliant solution for real-time arrhythmia detection, marking a step forward in deploying trustworthy artificial intelligence for decentralized medical diagnostics.