🌏 中文版
The PDF in the official HW6 ZIP is titled Homework 6: Learning Theory, MLE/MAP, Fairness Metrics, and Societal Impact; the coursework index shortens this to Learning Theory and Ethics. It is entirely written. Sections cover learning theory, MLE/MAP, probabilistic learning, fairness metrics, societal impacts and unintended consequences, and society/ethics/ML. The ZIP includes a PDF, LaTeX template, fairness CSV, and figures, but no starter code or reference answers.
One spine: why trust a model
Learning theory asks when a sample supports population claims. MLE and MAP expose the roles of data and priors. Fairness metrics ask how errors are distributed. None is a one-number verdict; assumptions, objectives, and affected groups come first.
First check and completion
Open the fairness_dataset.csv inside the ZIP and identify columns, groups, labels, and predictions before computing a metric. Completion means listing assumptions for every bound/estimator, reproducing fairness calculations from the CSV, and separating mathematically supported results from value judgments in societal-impact responses. There are no official answers to claim as verification.
Identify a fairness metric’s denominator
This is an invented teaching example, not an official assignment solution.
In invented group data, A has ten actual positives, eight predicted positive; B has five actual positives, four predicted positive. Both true-positive rates are 0.8. The denominator is actual positives, not the entire group population.
Check and answer: Does this prove all fairness conditions hold? No: without actual negatives, incorrect predictions, and group totals, false-positive rates or selection rates cannot be computed. Equal TPR establishes only equality on that metric.
Changelog
- 2026-10-11: Added an invented example, worked calculation or reasoning, and a comprehension check with an answer.
References
Loading...