As artificial intelligence (AI) becomes intrinsic to daily life, University of Florida (UF) researchers are prioritizing safe AI training methods. A groundbreaking study, “Deep Learning with Plausible Deniability,” co-authored by UF Ph.D. student Wenxuan Bao and associate professor Vincent Bindschaedler, Ph.D., addresses the significant issue of AI memorizing sensitive data, a growing privacy concern. Presented at NeurIPS 2025, this innovative approach incorporates a “privacy check” during AI training, ensuring that individual data points can’t be traced back to specific records. This technique leverages plausible deniability to prevent models from leaking crucial information like medical records, enhancing data privacy. Bindschaedler emphasizes the need for “trustworthy machine learning,” encompassing privacy, security, and interpretability. Acceptance at NeurIPS signifies UF’s increasing visibility in AI research, aligning with its commitment to advancing AI technology. Future research aims to expand this method’s applications and theoretical foundations, emphasizing its relevance in the evolving AI landscape.
Source link
Share
Read more