Jie Xu

Byzantine-robust federated learning can be sped up by orders of magnitude with provable convergence guarantees and minimal impact on error floors, simply by projecting gradients into a drastically smaller subspace before aggregation.

Differentially private federated learning gets a boost: PINA achieves 2.9% higher accuracy than state-of-the-art methods by using a novel two-stage approach with privacy-preserving initialization and normality-driven aggregation.

Current multimodal humor generation struggles to maintain image relevance, contextual appropriateness, and humor quality when explicitly conditioned on different cultural contexts, but this work offers a way to improve it.

Weakly supervised ViTs can achieve high accuracy in lymphoma classification, offering a practical alternative to fully supervised methods that require extensive manual annotation.