Uploaded July 2018 | Updated September 2026, 2 weeks ago
"Auditing, explaining, and ensuring fairness in algorithmic systems"
Machine learning models are becoming increasingly opaque to human examination, even to their designers. Yet these models are also increasingly used to make high-stakes decisions. In this talk, we’ll focus specifically on risk assessment models and research generated by a ProPublica examination of the failure modes of one recidivism prediction algorithm. This recent work from the field of Fairness, Accountability, and Transparency in machine learning includes a focus on the societal notions of fairness and non-discrimination. We will explain how these notions have been defined using a mathematical framework, discuss recently developed strategies for auditing black-box models when given access to their inputs and outputs, and consider methods for white-box interpretability in decision-making.
Jointly sponsored by the Data Science Institute and the Institute for Social and Economic Research and Policy
"Auditing, explaining, and ensuring fairness in algorithmic systems"
Machine learning models are becoming increasingly opaque to human examination, even to their designers. Yet these models are also increasingly used to make high-stakes decisions. In this talk, we’ll focus specifically on risk assessment models and research generated by a ProPublica examination of the failure modes of one recidivism prediction algorithm. This recent work from the field of Fairness, Accountability, and Transparency in machine learning includes a focus on the societal notions of fairness and non-discrimination. We will explain how these notions have been defined using a mathematical framework, discuss recently developed strategies for auditing black-box models when given access to their inputs and outputs, and consider methods for white-box interpretability in decision-making.
Jointly sponsored by the Data Science Institute and the Institute for Social and Economic Research and Policy
