Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Explore the theoretical foundations of bias amplification in machine learning through this seminar presented by Arjun Subramonian from Meta FAIR at USC Information Sciences Institute. Delve into how machine learning models capture and amplify biases present in training data, resulting in disparate performance across different social groups. Examine a precise analytical theory developed in the context of ridge regression, both with and without random projections, where the latter models feedforward neural networks in simplified regimes. Understand how model design choices and data distribution properties contribute to bias through rigorous mathematical frameworks. Gain insights into key phenomena including bias amplification and minority-group bias across various feature and parameter regimes. Learn about unified theoretical explanations that provide deeper understanding for evaluating and mitigating biases in machine learning systems. Discover research approaches that bridge theoretical analysis with practical fairness considerations in AI systems, presented by a researcher recognized with a FAccT 2023 Best Paper Award for contributions to machine learning fairness and ethics.
