Professor Saerom Park and her team in the Safe and Trustworthy AI (Safe AI) Lab within the Department of Industrial Engineering at UNIST have achieved a remarkable feat by securing the first-place position in the ‘CNN Interpretability’ category at the prestigious 2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML).
The IEEE SaTLM conference, sponsored by IEEE, focuses on machine learning security. A convolutional neural network (CNN) is a type of artificial neural network designed for processing and analyzing visual data, especially images.
The competition, held at the University of Toronto in Canada from April 9 to 11, aimed to advance discussions and innovations in machine learning security. It was the second time the competition took place since 2023, emphasizing new ideas and insights into the security and reliability of machine learning systems. This year’s theme centered on developing models that visualize triggers of backdoor attacks embedded in AI systems using cutting-edge visualization techniques.
Shown in the image is Researcher JiEun Yun from Professor Saerom Park’s Safe AI Lab, which secured 1st place in the ‘CNN Interpretability’ category at the IEEE SaTML 2024 Competition. Image Credit: Department of Industrial Engineering, UNIST
Under the guidance of Professor Park, a team consisting of JiEun Yun, YeSeong Jung, JonHyuk Backk, MinHwan Kim, and KiWan Kwon participated in the competition. Their model outperformed benchmarks, securing the top position. The team’s success can be attributed to their meticulous approach to adjusting hyperparameters and enhancing visualization performance to effectively identify and mitigate threats in AI models.
Researcher JiEun Yun stressed the importance of improving visualization performance to uncover hidden threats within AI models and expressed gratitude for the team’s hard work that led to international recognition. She remarked, “Detecting hidden threats in artificial intelligence models is crucial, and our victory underscores the significant enhancement in visualization performance through a simple yet innovative idea.” She further added, “This award holds great significance as our first international recognition after dedicated efforts from all team members. We are committed to furthering research in artificial intelligence security, an essential field in real-life applications.”
This award marks their first international recognition, highlighting their dedication to advancing research in artificial intelligence security, a critical field with practical applications.
