This course is directed to expose graduate students to newer facets of machine learning (ML). While ML literature has extensively focused on training accurate models (single objective), recent works have made some strides in lieu of multiobjective versions of the problem. For instance, problems in image recognition weigh a higher importance to recall (reduction of false negatives) over precision. Objectives like accuracy, computational complexity, and AUCROC are also equally important in any classification task, not restricted to imaging. Similarly, objectives like algorithmic bias (fairness) and explainability are very common in GDPR type applications involving a lot of personal data classification and handling. Our focus will be on understanding the current space of multiobjective ML literature.
- Course owner: Aswin Kannan