| layout | slideset | permalink |
|---|---|---|
handouts |
slides |
handout/ |
This week we will look at strategies for mitigating bias in machine learning and what their limitations might be. We'll conclude with a discussion about what other techniques might help and how things like interpretability, explanation, or data provenance might supplement algorithmic solutions.
Read these two papers: A comparative study of fairness-enhancing interventions in machine learning be prepared to critique the methodology of the evaluation. What additional things would you have done to evaluate the systems? What new questions do their results inspire?
Fairness and Abstraction in Sociotechnical Systems be prepared to generate concrete actions for a data scientist based on their recommendations for a more STS approach.
Read this blog post on IBM's implementation of FairML tools.
Optional, additional reading: Roles for computing in social change