Lab 8: Model Monitoring
In this lab, we'll add flagging to our ML-powered application so that users can give us feedback.
Then, we'll explore some data logged based on feedback from actual users of the FSDL Text Recognizer to the model monitoring and continual learning platform Gantry.
Outline
- 00:00 Basic user feedback with gradio
- 04:51 Logging feedback to Gantry
- 08:34 Checking for model toxicity with Gantry projections
- 14:23 Detecting model bugs in the Gantry UI with distributions and filters
- 19:01 Discovering surprising user data in the Gantry UI
- 29:53 Outro