ABSTRACT:

Deep learning models have matured to the point that high-performance model architectures are widely available and increasingly convergent, while approaches to engineering datasets have lagged. One modern solution is data-centric AI. This talk starts from an overview of data-centric AI, and then focuses on one specific direction: synthetic data generation. In particular, we present how artificially generated data can be used to preserve privacy, enhance fairness and increase adversarial robustness in several use cases. In the end, we highlight a list of challenges ahead of us in the era of data-centric AI.

BIO:

Guang Cheng is a Professor of Statistics at UCLA. Guang Cheng’s research interests are Trustworhty AI, Data-centric AI and Deep Learning Theory..

Hosted by Professor Quanquan Gu