Abstract

Foundation Model and Generative AI are among the most promising topics nowadays. However, their rapid developments also raise a series of critical challenges on responsibility such as privacy, security and Intellectual property (IP). In this talk, I will touch on the relevant risks and how we are championing the responsible development of Foundation Model and Generative AI, including using Federated Learning to empower the training/fine-tuning/adaptation of foundation models without touching on the original data; training models on public and synthetic data only; strategies to detect and address the unauthorized data usage, training data memorization, and IP infringements, ensuring compliance and trust.

Bio: Dr. Lingjuan Lyu is currently leading the privacy-preserving machine learning and vision foundation model team in Sony Research. Her main research interests include the low-cost foundation model and generative AI model development, responsible AI, and federated learning. She has won a series of awards, including AI 10 to Watch, IJCAI Early Career Spotlight, IEEE Outstanding Leadership Award, IBM Ph.D. Fellowship, National Scholarship, and Best/Outstanding/Oral/Spotlight paper awards or recognitions from top venues like ICML , ACL, CIKM , NeurIPS, AAAI , IJCAI, WWW , and KDD . She also served as a chair, committee member, or organizer of top conferences including her recent role as a chair for NeurIPS’24 Datasets and Benchmarks.