The effectiveness of today’s machine learning systems, e.g., image classification and text reasoning, highly hinges on the quality of their training data. In this talk, Lydia Y. Chen will present solutions on how to handle three types of data for (decentralized) deep learning systems, namely the bad noisy data, the ugly adversarial data, and the good synthetic data.
Översikt
- Datum:Startar 18 april 2023, 15:00Slutar 18 april 2023, 16:00
- Plats:EE hörsal, EDIT
- Språk:english
Lydia Y. Chen will first present novel generative adversarial networks (GANs) to generate the good synthetic data which not only resembles the original data but also serves as a means to share the data in a privacy-preserving manner. Then, she will introduce a robust learning framework to counter the noisy label data for single and multi-label classification models. Finally, she will conclude this talk by our recent adventure in adversarial learning. She will discuss the adversarial strategies employed by the malicious parties, which may aim to either lower the overall model accuracy or steal the deployed models.
Bio:
Lydia Y. Chen is the chair of the distributed systems section in the department of computer science at the Delft University of Technology in the Netherlands. Prior to joining TU Delft, she was a research staff member at the IBM Research Zurich Lab from 2007 to 2018.
She holds a PhD from Pennsylvania State University. Her research interests include distributed machine learning, dependability management, and large-scale data processing systems. She has published more than 100 papers in peer-reviewed journals and conference proceedings and received multiple best paper awards and nominations at flagship systems conferences. She received TU Delft technology fellowship in 2018.
She currently serves on the editorial boards of IEEE Trans. on Dependable and Secure Computing, IEEE Trans on Parallel and Distributed Systems, and IEEE Trans. on Network and Service Management. She is an IEEE senior member.
