My Fastai Course Note (3) Data Ethics

ifeelfree
Oct 30, 2020

--

The note is based on Fastbook.

  1. What is Ethics?

Well-founded standards of right and wrong that prescribe what humans ought to do

2. Bias

Bias Types

  • Historical bias: it comes from the fact that people are biased, processes are biased, and society is biased.
  • Measurement bias: it occurs when our models make mistakes because we are measuring the wrong thing, or measuring it in the wrong way, or incorporating that measurement into the model inappropriately.
  • Aggregation bias: it occurs when models do not aggregate data in a way that incorporates all of the appropriate factors.
  • Representation bias: the predicting model not only reflects the actual difference but also amplifies it.

Addressing Bias

Gathering a more diverse data set can address representation bias, but it will not help for other types of bias.

There is no such thing as a completely de-biased data set. Many researchers in the field have been converging on a set of proposals to enable better documentation of the decisions, context, and specifics about how and why a particular data set was created, what scenarios it is appropriate to use in, and what the limitations are. This way, those using a particular data set will not be caught off guard by its biases and limitations.

--

--

No responses yet