 |
Speaker: Qian Lou
Date: Nov 22, 2:05pm – 3:15pm Abstract: With growing reliance on cloud-based Machine Learning as a Service (MLaaS), privacy concerns are escalating, particularly in fields like healthcare and finance. For instance, clinicians and researchers use AI models to analyze electronic health records for insights into conditions such as depression and Alzheimer’s disease. However, increased MLaaS adoption also raises risks of data privacy violations, as seen in cases like the Facebook-Cambridge Analytica scandal and data leakage from Samsung employees using ChatGPT. To mitigate these risks, regulations such as the EU’s GDPR and similar laws in twelve U.S. states have been implemented, with more expected to follow. To ensure data security in privacy-sensitive fields, it is essential to develop robust, privacy-preserving frameworks for ML-based data analysis. Cryptography-based Private Computing (CryptoPC) methods, including homomorphic encryption (HE), oblivious transfer (OT), and secret sharing (SS), show substantial promise for secure MLaaS. However, despite their privacy advantages, CryptoPC-based private machine learning encounters efficiency barriers that limit widespread deployment. This talk will discuss the need for a co-design of cryptography and AI to create practical, efficient privacy-preserving AI solutions, addressing both the privacy and performance challenges in real-world MLaaS applications. Biographical Sketch: Qian Lou is an Assistant Professor of Computer Science at the University of Central Florida, specializing in privacy-preserving computing, secure machine learning, and computer systems. Formerly a Research Scientist at Samsung Research AI Center, he earned his M.S. and Ph.D. from Indiana University Bloomington, where he received the Luddy Outstanding Research Award. With over 40 publications across top venues like S&P, MICRO, and NeurIPS, his work has also earned recognition, including a Best Paper Nomination at ACM PACT. Location and Zoom link: 307 Love, or https://fsu.zoom.us/j/93566423235 |
|
