MLOps.community cover image

Cleanlab: Labeled Datasets that Correct Themselves Automatically // Curtis Northcutt // MLOps Coffee Sessions #105

MLOps.community

00:00

Understanding Epistemic and Alliteric Uncertainty in Models

In the process of model building, it's crucial to consider both epistemic uncertainty (intrinsic noise within the model) and alliteric uncertainty (noise from the data). Epistemic uncertainty stems from the model's inherent imperfections, where even with ideal data, 100% accuracy cannot be achieved. On the other hand, alliteric uncertainty arises from mislabeled or noisy data, causing the model to learn and replicate these errors. Distinguishing between epistemic and alliteric uncertainty requires making assumptions and leveraging techniques such as analyzing flipping rates among different classes. By addressing these uncertainties, like with the tool Clean Lab, one can enhance model performance and understand the model's behavior better, eventually improving it through collaborative code enhancement.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
 Get the app