top of page

🧙✨MyThisYourThat-for bias detection in federated learning published in Nature Digital Medicine!

Meet MyTH✨: Making bias detection a reality! 🧙‍♀️




MyThisYourThat is hot off the press at Nature Digital Medicine




Federated learning (FL) allows several data owners to collaboratively build AI models without sharing data.


Share models not data (see our related DISCO platform here 🕺🪩 https://lnkd.in/epMW9YHH )



While FL is great for privacy, it is not so great for transparency😶‍🌫️…How does one detect bias in data that we can’t see?


MyTH is a novel approach that adapts prototypical part learning networks to a federated learning setting. It enables each client in a federation to visualize feature differences learned by others on their own image: 


i.e. comparing one client’s 'This’ with others’ 'That’


 





Huge congratulations to Klavdiia Naumova, who did a PhD worth of meticulous work within an MSc thesis. This is certainly a name to keep an eye on 🌟👀



❤️Arnout Devos Sai Praneeth Karimireddy Martin Jaggi for your expert guidance and emojis.


bottom of page