The field of federated learning is rapidly advancing, with a focus on addressing the challenges of heterogeneous and imbalanced data, catastrophic forgetting, and privacy preservation. Recent developments have led to the creation of novel frameworks and methods that enable efficient and robust distributed learning, such as decentralized approaches and one-shot federated learning. These innovations have shown significant improvements in performance and scalability, making them suitable for real-world applications in medical imaging and healthcare. Notably, the integration of non-parametric models and the development of communication-efficient strategies have enhanced the accuracy and robustness of federated learning systems. Furthermore, researchers have made progress in addressing the issues of backdoor attacks, resolution drift, and Byzantine attacks, which are critical to the security and reliability of federated learning. Overall, the field is moving towards more efficient, scalable, and secure federated learning solutions that can be applied to various healthcare applications. Noteworthy papers include: VGS-ATD, which proposes a novel distributed learning framework that achieves high accuracy and scalability, and OptiGradTrust, which presents a comprehensive defense framework against Byzantine attacks and statistical heterogeneity. Additionally, FedS2R and FedCVD++ demonstrate the effectiveness of federated domain generalization and communication-efficient federated learning for synthetic-to-real semantic segmentation and cardiovascular risk prediction, respectively.
Advancements in Federated Learning for Medical Imaging and Healthcare
Sources
VGS-ATD: Robust Distributed Learning for Multi-Label Medical Image Classification Under Heterogeneous and Imbalanced Conditions
A New One-Shot Federated Learning Framework for Medical Imaging Classification with Feature-Guided Rectified Flow and Knowledge Distillation
Debunking Optimization Myths in Federated Learning for Medical Image Classification
FedS2R: One-Shot Federated Domain Generalization for Synthetic-to-Real Semantic Segmentation in Autonomous Driving
ModShift: Model Privacy via Designed Shifts
Guard-GBDT: Efficient Privacy-Preserving Approximated GBDT Training on Vertical Dataset
FedBAP: Backdoor Defense via Benign Adversarial Perturbation in Federated Learning
FedCVD++: Communication-Efficient Federated Learning for Cardiovascular Risk Prediction with Parametric and Non-Parametric Model Optimization
FLOSS: Federated Learning with Opt-Out and Straggler Support
Mitigating Resolution-Drift in Federated Learning: Case of Keypoint Detection
Seeing More with Less: Video Capsule Endoscopy with Multi-Task Learning
OptiGradTrust: Byzantine-Robust Federated Learning with Multi-Feature Gradient Analysis and Reinforcement Learning-Based Trust Weighting