# Report Outline
Bias assessment
Link to experiment grid
Link to statement of work

## Outline
1. Introduction
- Borrow from statement of work
- Introduce thesis
- Models implicitly learn bias from training data
- Result in biased output
- Introduce mitigation
- Train classifier to change predictions based on protected characteristic data in dataset
3. Background
-
5. Research question (hypothesis)
- Two research questions are explored in this project
- Is there a difference in the level of bias w.r.t. the statistical parity fairness metric on model output?
- Premise 1.A.: Models learn implicit bias from the dataset.
- Hypotheses:
- H₀: Models do not learn implicit bias.
- H₁: Models do learn implicit bias.
- Statistics:
- α = 0.05
- 1 - β = 0.80
- Test: two-sided t-test
- Is there a difference in the level of bias w.r.t. the model's architecture class?
- Hypothesis 1.B.: (i)
- Hypothesis 3: Architecture affects the amount of bias learnt by the model.
7. Research Design
8. Data Sets
9. Experiments
10. Results
11. Discussion
12. Conclusion
13. Recommendations