Johns Hopkins University
Title: Towards Collaborative, Intelligent, Fair Skin Disease Diagnostics with Differentially Private Federated Learning
Abstract: Early detection of skin lesions based on medical images can aid in identifying a range of infectious diseases with cutaneous manifestations. Lyme disease is an example of an infection with a potentially diagnostic skin lesion—which is caused by the bacterium Borrelia burgdorferi and causes nearly 476,000 cases per annum during 2010–2018. The talk on skin disease diagnostics has two parts. First, I will introduce our USENIX Security 2023 work, which studies the accuracy degradation cause of federated learning (FL) and Differential Privacy (DP) and then designs a differentially private, personalized data transformation to improve the accuracy. I will also discuss the possibility of using our work to achieve the practical goal of accurate, private diagnostics of skin disease diagnosis. Second, I will introduce our recent MICCAI 2023 work, which designs a novel preprocessing, data alteration method, called EdgeMixup, to improve model fairness with a linear combination of an input skin lesion image and a corresponding a predicted edge detection mask combined with color saturation alteration.
Biography: Dr. Yinzhi Cao is an assistant professor in Computer Science at Johns Hopkins University. His research mainly focuses on the security and privacy of the Web, smartphones, and machine learning using program analysis techniques. His past work was widely featured by over 30 media outlets, such as NSF Science Now (Episode 38), CCTV News, IEEE Spectrum, Yahoo! News, and ScienceDaily. He received three distinguished paper awards at USENIX Security'2021, SOSP’17, and IEEE CNS’15 respectively, and one best paper nomination at CCS’20. He is a recipient of the DARPA Young Faculty Award (YFA) 2022, the Amazon Research Award 2021 and 2017, and NSF CAREER Award 2021.
Host: Prof. Miaomiao Zhang
Organizers: Dr. Cong Shen and Dr. Mona Zebarjadi