|48’ watch||Meetup presentation: identify when and where bias can be introduced during ML process and how to minimize it|
Below is a talk I gave during the ML Fribourg meetup on November 27th 2019, based on a 2017 paper by Brian d'Alessandro, Cathy O'Neil and Tom Lagatta entitled Conscientious Classification: A Data Scientist's guide to Discrimination-Aware Classification.
The talk led to a fruitful discussion (not recorded) about the need for data normalization, the risk involved in augmenting the loss function with a fairness-regularizer, algorithm aversion, what incentives could companies have to introduce more fairness, the need for regulation and more.
You can download the slides here. They include additional links and a summary of the topics raised during the discussion.