Join the webinar "Using AI Techniques for Detection and Containment of Emerging Diseases" presented by Alberto Sangiovanni-Vincentelli from the University of California, Berkeley on Thursday, July 16 at 3:00 p.m. The presentation will be a lecture followed by a Q&A on distribution shift challenges in using AI to detect and contain COVID-19.
Abstract: When using AI methods to tackle the COVID-19 pandemic, distribution shifts from test data to training data create challenges. These shifts often lead to degraded performance, presenting obstacles when deploying machine learning models in healthcare AI and autonomous driving applications. In the first part of this talk, Professor Sangiovanni-Vincentelli will review the risks brought by incipient diseases and domain mismatch—two notable types of distribution shifts—to high stakes decision-making scenarios. To address the challenges, his research team has devised approaches that leverage the uncertainty information from ensemble learners and domain randomization. Their theoretical and empirical results show these approaches produce classifiers that are more robust against distribution shifts. For COVID-19 research, the sparsity of emerging disease data, especially at the initial outbreak of the epidemic, represents another challenge in using AI to detect and contain disease spread. In the second part of the talk, Professor Sangiovanni-Vincentelli will present his research team’s plan for tackling this challenge by building upon the above-mentioned techniques and incorporating them into state-of-the-art, human-AI collaborative methods. This is a joint investigation with Geoff Tison of UCSF and Yuxin Chen of the University of Chicago.
Register for the Zoom webinar
