From the course: Responsible AI with Amazon SageMaker AI

Unlock this course with a free trial

Join today to access over 25,600 courses taught by industry experts.

Detecting bias in post-training and production

Detecting bias in post-training and production - Amazon SageMaker Tutorial

From the course: Responsible AI with Amazon SageMaker AI

Start my 1-month free trial Buy for my team

Detecting bias in post-training and production

“

- [Instructor] Did you know that even the most well-trained models can develop bias after deployment? It's called post-training bias, and it happens when models encounter real-world data that wasn't in the training set. Left unchecked, it can lead to unfair outcomes. You can use SageMaker Clarify to detect and monitor bias in post-training and production models. SageMaker Clarify provides 11 post-training metrics to reveal bias and fairness. Let's explore two of the most popular metrics, disparate impact and difference in positive proportions in predicted labels, or DPPL. You can review the handout to explore the remaining metrics. Disparate impact compares the rate of positive predictions, so yes to a hiring decision or loan approval between groups. Disparate impact answers questions like, are different demographic groups receiving positive predictions at similar rates? DPPL measures the difference in the proportion of positive predictions between groups. DPPL answers questions like,…

Contents