Embedded Files
Bridging Data Silos with practical federated learning
Bridging Data Silos with practical federated learning
Abstract: Critical sectors such as healthcare, finance, and public services often have the data needed for high-impact AI, but that data is fragmented across organizations and constrained by privacy, regulation, and operational barriers. Federated learning enables multiple parties to train models collaboratively without centralizing raw data—yet real deployments demand more than accuracy. This talk discusses how to enable cross-silo federated learning in practice: combining rigorous privacy mechanism with heterogeneous data and unreliable communication, defending poisoned data from clients, and enabling machine unlearning when data removal is needed. It also discusses future directions for federated learning, highlighting the challenges and opportunities in scaling trustworthy deployment for critical applications.
Dr. Zhaomin Wu (National university of Singapore)
Dr. Zhaomin Wu (National university of Singapore)
Zhaomin Wu is a Research Fellow at the Department of Computer Science, National University of Singapore. His research focuses on trustworthy machine learning, with special interests in federated learning, machine unlearning, and trustworthy LLMs. His publications appear in top-tier venues including NeurIPS, ICLR, SIGMOD, KDD, ACL, EMNLP, AAAI, MLSys, and TKDE. In 2023, He received SIGMOD Honorable Mention for Best Artifact Award.
Verifiable Federated Learning: A TEE-Based Approach for Next-Generation Privacy Guarantees
Verifiable Federated Learning: A TEE-Based Approach for Next-Generation Privacy Guarantees
Abstract: Google’s production system for federated learning now leverages trusted execution environments (TEEs) to address some of the challenges of cross-device federated learning. The system offers full external verifiability of the server-side components of federated learning, improves operability, and enables scaling to much larger models. In this talk, we’ll explain the history and evolution of FL at Google, and introduce an updated definition of federated learning based on its privacy principles (transparency/auditability, data minimization, and data anonymization) rather than on the placement of data processing. We’ll describe how the new approach compares to traditional cross-device federated learning and some new algorithms and use cases unique to the TEE-hosted federated learning setting.
Katharine Daly (Google)
Katharine Daly (Google)
Katharine Daly has built infrastructure for multiple generations of federated learning and federated analytics systems at Google Research. Recently she has focused on designing scalable systems that achieve verifiable differential privacy guarantees via TEEs (Trusted Execution Environments) for GenAI use cases.
Dr. Daniel Ramage (Google)
Dr. Daniel Ramage (Google)
Daniel Ramage directs the Google Research teams responsible for the production systems and research roadmap powering federated learning at Google. He is a co-inventor of federated learning and federated analytics, overseen their deployment in Google systems, and focuses on systems and methods for private and secure AI.
Federated Learning enables Collaborative Model Development for Medical Imaging
Federated Learning enables Collaborative Model Development for Medical Imaging
Abstract: The application of artificial intelligence in medical imaging offers substantial potential to advance clinical practice. However, the development of effective models is often constrained by the isolated nature of sensitive imaging data, which remains siloed across institutions due to privacy and regulatory concerns. Federated learning presents a transformative approach by facilitating collaborative model development across multiple data sources without centralizing the data itself. This talk will discuss the paradigm of federated learning in the context of medical imaging. It outlines the key challenges and highlights directions aimed at making federated learning more efficient, robust, and applicable for advancing the field of medical imaging analysis.
Prof. Qi Dou (The chinese university of hong kong)
Prof. Qi Dou (The chinese university of hong kong)
Prof. Qi Dou is currently an Associate Professor at the Department of Computer Science and Engineering at The Chinese University of Hong Kong (CUHK). She is an affiliated member of CUHK T-Stone Robotics Institute, CUHK Institute of Medical Intelligence and XR, Hong Kong Multi-scale Medical Robotics Center, and Hong Kong Centre for Logistics Robotics. Her research is in the interdisciplinary field of AI and robotics technologies for medical applications including autonomous surgical robot, medical imaging, safe embodied AI, etc. Her publications have received a number of prestigious best paper awards and nominations, including IJCARS-MICCAI Best Paper Awards (2021 & 2024), MICCAI Best Paper Award Finalists (2023 & 2024), MICCAI Young Scientist Publication Award (2022), IEEE ICRA Best Paper Award in Medical Robotics (2021), IEEE TBME Prize Paper Award (2021), MedIA-MICCAI Best Paper Award (2017). She serves as General Co-Chair for MICCAI 2026, Program Co-Chair for MICCAI 2024, IPCAI 2023, MICCAI 2022, and MIDL 2022, and Area Chairs for NeurIPS, CVPR, AAAI, ICRA, IROS, etc.
Page updated
Google Sites
Report abuse