Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the fundamental tensions between different fairness criteria in machine learning through this 59-minute conference talk that examines why error-rate parity and calibration within subgroups typically cannot be achieved simultaneously. Learn about a novel approach to relaxing these fairness notions that illuminates the inherent tradeoffs between them and discover a spectrum of optimally fair prediction rules that emerge from this framework. Understand when certain fairness requirements are mathematically impossible to satisfy and delve into the complex question of how to construct fair decision-making processes from fair scoring systems. Gain insights into the theoretical foundations of algorithmic fairness and the practical implications of these limitations for designing equitable machine learning systems in healthcare and other domains.
