| Title | : | Machine Unlearning |
| Speaker | : | Prof. Mohan Kankanhalli (School of Computing, NUS) |
| Details | : | Fri, 17 Nov, 2023 11:00 AM @ SSB 334 |
| Abstract: | : | Big data has been one of the important enablers of the recent advances in Artificial Intelligence. Its use for training machine learning models has enabled many useful applications. However, there are also some downsides when the data contains sensitive, often personal, information. Recent privacy and data protection regulations across the world have recognized the “right to be forgotten†as a fundamental right of citizens. Enabling this in practice not only requires deletion of data from institutional databases but also necessitates deletion of data from trained machine learning models. Machine Unlearning refers to the process of selectively removing or forgetting some specific data or a class of data from a machine learning (often deep learning) model. This talk will introduce the emerging field of machine unlearning, discuss the motivations behind it, the techniques employed, and its implications for data privacy, fairness, and interpretability. We will present several approaches for unlearning, including the selective modification of model parameters, regularization techniques, data-free methods, and teacher-student frameworks. We will then discuss the challenges and open research questions in the field. Machine Unlearning can thus be seen to contribute towards building more accountable AI systems that can engender trust among users and stakeholders. |
