Guang Lin receives NSF grant to study deep learning in real physical space
“The vulnerability of deep neural networks to small and imperceptible perturbations is a major challenge in machine learning today,” said Guang Lin, professor of mathematics and mechanical engineering. “Existing theoretical studies, while laying a good foundation based on advanced statistical analyses, require various idealistic assumptions. In order to truly test the robustness of deep learning algorithms, they need to observed in a real physical environment.”
To this end, Lin has recently received a 3-year, $1 million grant from the National Science Foundation to research “Robust Deep Learning in Real Physical Space: Generalization, Credibility, and Scalability.” Lin is the principal investigator, and will be joined by co-principal investigators Stanley Chan, Elmore Associate Professor of Electrical and Computer Engineering and associate professor of statistics; Guang Cheng, professor of statistics; Jean Honorio, assistant professor of computer science; and Yian Ma, assistant professor at the Halcolu Data Science Institute at the University of California San Diego.
The team has chosen computer vision – computers recognizing and classifying visual images – as a focus of their work. They plan on developing their theories and models, and then building a computational photography testbed to implement their concepts and validate the theoretical results. By co-modeling both the deep neural networks and the environment in which the neural networks are operating, they hope to close the gap between theory and reality.
The prototype optical system was initially started with Professor Chan and Professor Garth Simpson in 2018. Three students - Alex Sherman, Casey Smith and Xiangyu Qu - made the first attempt to build the optical system. In 2019, Professor Chan received a seed grant from Purdue University College of Engineering, with the help of Karthik Ramani. In 2019, Professor Chan got support from the Army to develop a different approach. In Fall 2020, Professor Chan and his student Abhiram Gnanasambandam finally built the system, and published a corresponding paper at the ICCV 2021 workshop. This work has created a huge interest among the community: https://cacm.acm.org/news/254919-optical-adversarial-attack-can-change-the-meaning-of-road-signs/fulltext
“We have a great team,” said Lin. “The combination of skills we have in mechanical engineering, electrical engineering, statistics, and computer science offer a unique opportunity to address this complex problem.”
Writer: Jared Pike, firstname.lastname@example.org, 765-496-0374
Source: Guang Lin, email@example.com