Google Model Cards: A Proposed First Step

Author: N/A

Publisher: Google Cloud

Publication Year: N/A

Summary: The following tool is a proposed method for machine learning transparency similar to Microsoft’s Datasheets for Datasets. It proposes a “card” that comes with each model, designed for experts and non-experts, with information pertaining to the content, process, experience, fairness, and privacy of each model. There are sample cards given for their face and object detection functions of their Cloud Vision API. These example cards give the overview, limitations, indicators of performance, and other information on the cards that still be understood by non-technical audiences. There is also an accompanying white paper detailing specific recommendations and items to include depending on the type of analysis and methods used for the model.


Posted

in

, ,

by