Introduction
- There is potential for machine learning models to cause harm.
- Researchers are increasingly required to reflect on the impact of their work.
- Regulation and oversight are in their infancy.
Tasks
- Not all applications of machine learning are for the public good.
Data
- Data is fundamental to the field of machine learning.
- Datasheets can help us to reflect on the process of data creation and distribution.
Fairness
- Biases in data lead to biased models.
- All current models are likely to exhibit some form of bias.
- Achieving fairness is an increasingly active area of research.
Dataset shift
- Dataset shift can result from changes in technology, population, and behaviour.
- Dataset shift can lead to deterioration of models after deployment.
- Dataset shift is a major issue in terms of deployment of machine learning models.
Explainability
- “The importance of explainability is a matter of debate.”
- “Saliency maps highlight regions of data that most strongly contributed to a decision.”
Attacks
- Models are susceptible to manipulation.