Timezone: »

GenLabel: Mixup Relabeling using Generative Models
Jy yong Sohn · Liang Shang · Hongxu Chen · Jaekyun Moon · Dimitris Papailiopoulos · Kangwook Lee

Wed Jul 20 07:35 AM -- 07:40 AM (PDT) @ Ballroom 3 & 4

Mixup is a data augmentation method that generates new data points by mixing a pair of input data. While mixup generally improves the prediction performance, it sometimes degrades the performance. In this paper, we first identify the main causes of this phenomenon by theoretically and empirically analyzing the mixup algorithm. To resolve this, we propose GenLabel, a simple yet effective relabeling algorithm designed for mixup. In particular, GenLabel helps the mixup algorithm correctly label mixup samples by learning the class-conditional data distribution using generative models. Via theoretical and empirical analysis, we show that mixup, when used together with GenLabel, can effectively resolve the aforementioned phenomenon, improving the accuracy of mixup-trained model.

Author Information

Jy yong Sohn (University of Wisconsin-Madison)
Liang Shang (University of Wisconsin-Madison)
Hongxu Chen (University of Wisconsin-Madison)
Jaekyun Moon (KAIST)
Dimitris Papailiopoulos (University of Wisconsin-Madison)
Kangwook Lee (UW Madison)

I am an Assistant Professor at the Electrical and Computer Engineering department and the Computer Sciences department (by courtesy) at the University of Wisconsin-Madison. Previously, I was a Research Assistant Professor at Information and Electronics Research Institute of KAIST, working with Prof. Changho Suh. Before that, I was a postdoctoral scholar at the same institute. I received my PhD in May 2016 from the Electrical Engineering and Computer Science department at UC Berkeley and my Master of Science degree from the same department in December 2012, both under the supervision of Prof. Kannan Ramchandran. I was a member of Berkeley Laboratory of Information and System Sciences (BLISS, aka Wireless Foundation) and BASiCS Group. I received my Bachelor of Science degree in Electrical Engineering from Korea Advanced Institute of Science and Technology (KAIST) in May 2010.

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors