November 10, 2020

ECE student earns Honorable Mention for CRA Outstanding Undergraduate Researcher Award

The Computing Research Association has awarded Purdue ECE's Seo Young Kim an Honorable mention for the CRA Outstanding Undergraduate Researcher Award for 2020. Kim earned her bachelor’s degree in Computer Engineering in May 2020.
Seo Young Kim
Seo Young Kim, BS CmpE '20

The Computing Research Association has awarded Purdue ECE's Seo Young Kim an Honorable mention for the CRA Outstanding Undergraduate Researcher Award for 2020. Kim earned her bachelor’s degree in Computer Engineering in May 2020. The CRA award program recognizes undergraduate students in North American universities who show outstanding potential in an area of computing research. Kim says she is very proud of herself for earning an honorable mention.

“Research was the reason why I decided to pursue graduate studies,” says Kim, “so this award means a lot to me personally as receiving the honorable mention means that my research work has been recognized.”

Kim’s research was in the area of computer vision. She says computer vision, especially image classification, has achieved impressive progress in recent years, but a major weakness of image classification is the dependency on training datasets with the scarcity of labeled data. Kim led a research team analyzing bias between different datasets by determining which images in datasets are similar and therefore improving transferability for training. Her team determined the correlation between the measured similarities between datasets and classification results for various train-test combinations in order to answer the following question: Out of the various source datasets that are available, how should one select the best dataset to maximize the model performance? Additionally, the team proposed a pseudo-labeling method to obtain labels without human efforts after choosing a better dataset to train. The method reduces the dependency on labels and source dataset while improving model performance without any labels from the target dataset.

Share