Fairness in face recognition in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What does fairness mean in face recognition systems?

Fairness means the system works equally well for all groups of people, regardless of race, gender, or age, avoiding bias or discrimination.

beginner

Why can face recognition systems be unfair?

They can be unfair because training data might have more images of some groups than others, causing the system to perform worse on underrepresented groups.

intermediate

What is demographic parity in face recognition fairness?

Demographic parity means the system's positive prediction rates are similar across different demographic groups, like race or gender, ensuring no group is unfairly treated.

intermediate

How can we reduce bias in face recognition models?

We can reduce bias by using balanced datasets, applying fairness-aware training methods, and testing the model on diverse groups to check performance.

beginner

What is the impact of unfair face recognition systems in real life?

Unfair systems can lead to wrongful identification, privacy violations, and discrimination, affecting people's trust and safety.

What is a common cause of unfairness in face recognition systems?

A High-resolution images

B Using too many layers in the model

C Imbalanced training data

D Using grayscale images

Imbalanced training data causes the model to perform better on groups with more data, leading to unfairness.

Which fairness metric checks if error rates are similar across groups?

A Demographic parity

B Accuracy

C Precision

D Recall

Demographic parity ensures error rates are balanced across different demographic groups.

What is one way to test fairness in a face recognition model?

A Test only on the training data

B Test on diverse demographic groups

C Use only one demographic group for testing

D Ignore testing and deploy immediately

Testing on diverse groups helps identify if the model is fair or biased.

Why is fairness important in face recognition?

A To improve model speed

B To increase training time

C To reduce image size

D To avoid discrimination and errors

Fairness helps prevent discrimination and wrongful identification.

Which of these can help reduce bias in face recognition?

A Balanced datasets and fairness-aware training

B Ignoring minority groups

C Using biased datasets

D Reducing model complexity only

Balanced datasets and fairness-aware training help reduce bias.

Explain why fairness is a challenge in face recognition systems and how it can affect different groups.

Think about how training data and model errors relate to fairness.

Describe methods to improve fairness in face recognition models.

Consider both data and model training approaches.