The current developments in the research area are significantly advancing the integration of privacy-preserving techniques with machine learning applications, particularly in healthcare and urban environments. Federated learning (FL) is emerging as a cornerstone, enabling collaborative model training across multiple institutions without compromising data privacy. Innovations in FL are being complemented by advancements in graph neural networks (GNNs) and differential privacy (DP), enhancing the ability to analyze complex data types such as EEG signals and spatio-temporal data while maintaining robust privacy protections. Notably, the field is witnessing a shift towards more efficient and verifiable secure multi-party computing protocols, which are crucial for privacy-preserving machine learning. These protocols are not only improving computational efficiency but also ensuring higher accuracy and verifiability in secure matrix operations, which are foundational for various machine learning tasks. The integration of these technologies is paving the way for more accurate and explainable predictive models in critical areas such as stroke assessment and urban infrastructure enhancement, while also addressing the pressing need for privacy and security in data-sharing environments.

Noteworthy papers include one that introduces a novel federated learning method combined with GNNs for stroke severity prediction using EEG signals, demonstrating high accuracy and explainability while preserving data privacy. Another notable contribution is a study that proposes an efficient, verifiable, and accurate secure three-party computing framework for matrix operations, significantly reducing communication overhead and enhancing precision in secure computations.