Federated Learning Advances

The field of federated learning is moving towards addressing key challenges in scalability, privacy, and data heterogeneity. Researchers are exploring innovative methods to improve the accuracy and efficiency of federated learning models, including new client selection algorithms, secure aggregation methods, and knowledge distillation techniques. Noteworthy papers in this area include the proposal of Knowledgeable Client Insertion, which introduces a small number of knowledgeable clients to improve learning accuracy, and the development of an adaptive clustering scheme for client selections, which dynamically adjusts the number of clusters to reduce communication costs. Another notable work is the demonstration of training a neural network using fully homomorphic encryption, enabling privacy-preserving and efficient communication. Additionally, the proposal of a novel client-level assessment of collaborative backdoor poisoning in non-IID federated learning highlights the importance of addressing security vulnerabilities in federated learning scenarios.

Sources

The More is not the Merrier: Investigating the Effect of Client Size on Federated Learning

An Adaptive Clustering Scheme for Client Selections in Communication-Efficient Federated Learning

Optimising Intrusion Detection Systems in Cloud-Edge Continuum with Knowledge Distillation for Privacy-Preserving and Efficient Communication

Divergence of Empirical Neural Tangent Kernel in Classification Problems

Diversity-Driven Learning: Tackling Spurious Correlations and Data Heterogeneity in Federated Models

Local Data Quantity-Aware Weighted Averaging for Federated Learning with Dishonest Clients

Stochastic Gradient Descent in Non-Convex Problems: Asymptotic Convergence with Relaxed Step-Size via Stopping Time Methods

Privacy-Preserving CNN Training with Transfer Learning: Two Hidden Layers

Decentralized Nonconvex Composite Federated Learning with Gradient Tracking and Momentum

A Client-level Assessment of Collaborative Backdoor Poisoning in Non-IID Federated Learning