# Site

## Pages

- [Page not found](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/404/): That page doesn't exist, or it has moved since you last looked
- [Assessment](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/assessments/): Every piece of graded work in the course, what it is worth, and when it is due.
- [Assignment 1: Reproduce the number](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/assessments/a1-reproduce-the-number/): Take a published FID from a real paper. Reproduce it. Explain the gap. Marked on the explanation, not on the size of the gap.
- [Assignment 2: Break the number](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/assessments/a2-break-the-number/): Construct a bench candidate that scores well and is visibly wrong, or scores badly and is visibly fine. Submit the set, the score and the mechanism.
- [Final report](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/assessments/final-report/): Propose what you would report in place of the score, then evaluate your proposal by the standard this course applied to FID. Most proposals fail. Say how yours does.
- [Measurement log](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/assessments/measurement-log/): One entry a week from week 1. Every entry records a number, the sample size it was computed at, and the implementation that produced it.
- [Lectures](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/): A live introduction, four formal lectures and a course closing: derive the number, question the estimate, change the instrument, defend the report.
- [The bench, live](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-01/): The week 1 exercise as an instrument you can run in this page. Draw your own samples, score them, and watch the estimate disagree with the truth.
- [The Fréchet distance](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-02/): The week 2 derivation delivered as a sequence rather than a page, one step of the chain per slide, ending on the bench scored by hand
- [Question the estimate](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-05/): Bias and noise are different problems. Return to the week 2 baseline, measure the bias on the bench, and take its price with it.
- [Change the instrument](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-08/): Two honest numbers from the same recordings can disagree about which candidate is worse. Put three instruments on one bench and find out why.
- [Defend the report](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-11/): A number is not evidence. Declare the protocol behind a claim, run it twelve times, and find out what it will carry.
- [The number is not the conclusion](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/lectures/week-12/): A careful account of what a score supports is worth more than a confident ranking. A closing word before you finish the final report.
- [People](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/people/): Who teaches the course, and how to reach them.
- [Dr Helen Sandoval](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/people/helen-sandoval/): Convenor and lecturer. Leads the Measurement Reliability Group, studying finite-sample bias and what evidence a reported model ranking needs.
- [Tomasz Wierzba](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/people/tomasz-wierzba/): PhD candidate, tutor and bench maintainer. Studies reproducible evaluation pipelines and helps students separate sampling variation from implementation errors.
- [Policies and support](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/policies/): The course-wide rules for late work, reporting, generative tools and academic integrity. Every assessment brief links here rather than restating them.
- [The number](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/01-the-number/): The bold number in a results table is the field's account of what got better, and nothing in the table says what it was computed against. The bench is built here.
- [The Fréchet distance](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/02-frechet-distance/): The closed form for two Gaussians, derived rather than quoted, and the assumption it smuggles in
- [The instrument](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/03-the-instrument/): Inception-v3 was trained to name a thousand ImageNet classes. The score measures distance in the space it built to do that.
- [Getting it right](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/04-getting-it-right/): The same samples can receive different scores. Separate preprocessing from numerical arithmetic, then pin the protocol you can defend.
- [The bias](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/05-the-bias/): Finite-sample FID is biased, the bias depends on the generator, and fixing N does not repair the comparison
- [Blind spots](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/06-blind-spots/): A generator that memorised the reference set scores perfectly. What else the score cannot see, demonstrated on the bench.
- [Dropping the Gaussian](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/07-dropping-the-gaussian/): An unbiased estimator need not detect every difference. Compare KID, density and coverage against the bench's known Gaussian-distance baseline.
- [Video](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/08-video/): FVD keeps the Gaussian distance but changes the features. Test what a temporal instrument preserves before transferring claims from images.
- [Human evaluation](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/09-human-evaluation/): The judgement the score stands in for. How a rating study is designed, what it costs, and why it is so rarely run.
- [Correlation](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/10-correlation/): Put the ranking a score produces beside the ranking people produce, and ask whether they agree. That question decides whether the previous nine weeks were worth doing.
- [Why it persists](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/11-why-it-persists/): Nine years of results tables carry an FID column, and that continuity is the strongest argument anyone has for keeping it. What the number costs to leave out, stated honestly.
- [What you would report instead](https://comp4020-agentic-coding-studio.github.io/comp4020-ass2-jnheinrich451-eng/sessions/12-what-you-would-report-instead/): Defend your proposal by the standard built across the semester. Bring the final bench comparison and the conditions under which you would still report FID.