Description:
Privacy-preserving AI over Next Generation Networks
Princeton Docket # 24-4056-1
Artificial intelligence (AI) is a key technology for next-generation networks; it will empower new functionalities to enable low-latency inference and sensing applications, including autonomous driving, personal identification, and activity classification (to name a few). Two conventional AI paradigms are commonly used in practice: 1) On-device inference which locally performs AI-based inference, which however suffers from high computation overhead, and 2) On-server inference, where edge devices upload their raw data to a central server to perform a global inference task; however, this framework violates the data privacy of individuals, and it suffers from high communication overhead. To remedy these challenges, edge-device collaborative inference is a compelling solution. In this setting, the joint inference is divided into three modules: a) sensing for data acquisition, b) feature extraction, and 3) feature encoding for transmission. Obtaining fine-grained information about individuals is a risk that must be considered while designing such task-oriented communication systems.
To address this challenge, Princeton researchers have developed a novel private collaborative inference mechanism wherein each edge device protects the sensitive information of the extracted features before transmission to the central server for inference. The key design objectives of this invention are two-fold: 1) minimize the communication overhead and 2) maintain rigorous privacy guarantees for transmission of features over communication networks while providing satisfactory inference performance.
Applications
• Many use cases over different communication networks (e.g., online social networks, financial networks, indoor Wi-Fi, mobile networks)
• Wide range of applications over such networks (e.g., anomaly detection, traffic monitoring, financial trading, portfolio optimization, distributed sensing, object detection, speech recognition, activity classification)
• Example use case: Traffic Prediction for Smart cities
Advantages
• Enhanced privacy: system protects sensitive information by obfuscating private features before transmission
• Reduced communication overhead: by extracting and encoding only relevant features, the system minimizes data transmission requirements
• Flexible privacy levels: framework allows for precise calibration of privacy measures to suit specific needs and contexts
• Efficient edge processing: system enables integration of privacy mechanisms using pre-trained ML models, without extensive redesign
• Versatility: framework supports various applications across many different network types (e.g., anomaly detection, distributed sensing)
Stage of Development
Researchers have completed a proof-of-concept study of this approach using an inference system with two edge devices and one edge server for performing two-view image classification based on the MNIST dataset.
Publication
[2410.19917] Collaborative Inference over Wireless Channels with Feature Differential Privacy
Inventors
Mohamed Seif, Ph.D. is a postdoctoral fellow in the Department of Electrical and Computer Engineering at Princeton University, advised by Andrea Goldsmith and H. Vincent Poor. He completed his Ph.D. in Electrical and Computer Engineering at the University of Arizona.
Andrea Goldsmith, Ph.D. is the Dean of Engineering and Applied Science and the Arthur LeGrand Doty Professor of Electrical and Computer Engineering at Princeton University.
Her research interests are in information theory, communication theory, and signal processing, and their application to wireless communications, interconnected systems, and neuroscience. She founded and served as Chief Technical Officer of Plume WiFi (formerly Accelera, Inc.) and of Quantenna (QTNA), Inc, and she currently serves on the Board of Directors for Medtronic (MDT) and Crown Castle Inc (CCI). Dr. Goldsmith is a member of the National Academy of Engineering and the American Academy of Arts and Sciences, a Fellow of the IEEE and of Stanford, and has received several awards for her work, including the IEEE Sumner Technical Field Award, the ACM Athena Lecturer Award, the ComSoc Armstrong Technical Achievement Award, the Kirchmayer Graduate Teaching Award, the WICE Mentoring Award, and the Silicon Valley/San Jose Business Journal’s Women of Influence Award.
She is the author of the book "Wireless Communications'' and co-author of the books "MIMO Wireless Communications'' and “Principles of Cognitive Radio,” all published by Cambridge University Press, as well as an inventor on 29 patents. She received B.S., M.S. and Ph.D. degrees in Electrical Engineering from U.C. Berkeley.
Dr. Goldsmith is currently the founding Chair of the IEEE Board of Directors Committee on Diversity, Inclusion, and Ethics. She served as President of the IEEE Information Theory Society in 2009, as founding Chair of its Student Committee, and as founding Editor-in-Chief of the IEEE Journal on Selected Areas of Information Theory.
H. Vincent Poor is the Michael Henry Strater University Professor of Electrical Engineering at Princeton University. Dr. Poor is widely recognized as one of the world's leading educators and researchers in wireless communications, signal processing and related fields. Dr. Poor's graduate-level textbook, "An Introduction to Signal Detection and Estimation," is considered the definitive reference in this field.
His current research activities are focused on advances in several fields of rapid technology development, notably wireless networks and energy systems, and on the fundamentals underlying them, including information theory, machine learning and network science.
In recognition of his contributions and innovative research, Dr. Poor has received numerous awards and distinctions, including the IEEE Alexander Graham Bell Medal and induction into the National Academy of Engineering and the National Academy of Sciences.
Intellectual Property & Development status
Patent protection is pending.
Princeton is currently seeking commercial partners for the further development and commercialization of this opportunity.
Contact
Renee Sanchez, JD
Princeton University Tech Licensing & New Ventures • Phone: (609) 258 6762 • Email: rs1453@princeton.edu
