CPI Talk: Nicolas Papernot on Characterizing Machine Unlearning through Definitions and Implementations
CPI would like to thank Nicolas Papernot, assistant professor of computer engineering and computer science at the University of Toronto, for coming out to the University of Waterloo and giving such an engaging talk on Thursday, June 20. Papernot presented the benefits of machine unlearning along with its challenges and potential solutions. For example, he shared it is not always clear which data points one wants to unlearn. However, utilizing differential privacy can help to address the compounding effect (unlearning old ideas) that eventually leads to the creation of new ideas.
Papernot also discussed that one of the problems with unlearning is that it is tied to generalization and urged the collaboration of security and privacy experts with machine learning “friends” to further enhance research development in this area.
For more on machine unlearning, check out CLEVERHANS-BLOG or watch the video recording below.