It is a novel approach that uses attention maps from GNNs to distill structural information from the original dataset into synthetic graphs.

We introduce a novel personalized federated learning algorithm that leverages the inherent graph-based relationships among clients.

a method for handling noisy labels in text-to-image person re-identification that outperforms previous approaches by using dual noise detection and triplet loss.

We introduce a novel personalized federated learning algorithm that leverages the inherent graph-based relationships among clients.

We introduce PerSOTA FL (Per-Sonalized Over-the-Air Federated Learning), utilizing QR decomposition for robust OTA-based aggregation, particularly effective in low SNR environments.

We introduce a novel deep learning-based approach for denoising Electrocardiogram (ECG) signals by leveraging the periodicity of cardiac cycles through a convolutional neural network (CNN) model equipped with a local/non-local cycle observation (LNC) module.

We introduce a deep-learning framework, to accurately extract maternal and fetal ECG from single-channel abdominal recordings using two-stage deep model.

Introducing "PnP-AugHQS," an enhanced PnP-HQS technique for sparse recovery, featuring comprehensive convergence analysis and a compact deep CNN denoising regularization for improved performance, adaptable for MRI reconstruction.

We tackle device selection and resource allocation in an indoor setting with multiple smart devices using hybrid VLC/RF communication, employing multi-objective optimization and intelligent device prioritization based on local gradients alignment.

Introducing a novel VLC-based Intelligent Transportation System (ITS) utilizing vehicle LED headlamps as remote controllers in a practical scenario.

Introducing a framework for robust contactless heart rate (HR) estimation from facial videos, leveraging diverse color space representations and PPG signal extraction methods. The framework incorporates a novel PPG amplitude variation elimination technique (AVET) to enhance signals, followed by fusion algorithms for improved HR estimation accuracy.

We introduce the Arithmetic Shifter-Based Support Vector Machine (ASB-SVM) classifier, leveraging shift registers for faster signal classification, achieving 98.2% accuracy for green modulated signals in CRNs with significant resource reduction and speed increase.