The recent developments in the research area of privacy-preserving data analysis and federated learning have shown a strong focus on enhancing data privacy and security while maintaining computational efficiency and model utility. Researchers are increasingly exploring innovative methods to integrate differential privacy (DP) into federated learning (FL) frameworks, addressing the challenges of data confidentiality and the 'right to be forgotten.' Notably, advancements in federated unlearning have introduced novel strategies to ensure that models can effectively 'forget' specific data points without compromising overall model performance. Additionally, there is a growing interest in developing scalable systems that protect the privacy of individual users while enabling collaborative machine learning, leveraging trusted execution environments (TEEs) to ensure data integrity and confidentiality. Furthermore, the integration of DP mechanisms into on-device analytics has opened new avenues for processing ephemeral data streams securely, with applications in sustainability and beyond. These developments collectively push the boundaries of privacy-preserving technologies, offering practical solutions for real-world data challenges.

Noteworthy papers include one proposing Federated Unlearning with Indistinguishability (FUI) for DPFL, which introduces a novel approach to unlearning data while maintaining model indistinguishability. Another highlights Mayfly, a federated analytics system that enables private aggregate insights from ephemeral streams of on-device user data, showcasing its application in sustainability.