AWS Certified AI Practitioner
Datasets, Bias, and Fairness
Where bias enters a model through its training data, what makes a dataset inclusive and balanced, and why the word bias means two different things on this exam.
Intermediate 22 minutes 5 Learning Objectives
- Identify the ways bias enters a model through its training data
- Describe the characteristics of a good dataset: inclusivity, diversity, curated sources, and balance
- Distinguish statistical bias and variance from societal bias, and connect them to underfitting and overfitting
- Explain why competing definitions of fairness cannot all be satisfied at once
- Select dataset-stage mitigations for a described bias problem