to 
Basically they’re investigating how to improve smoothness constraints by incorporating data, tasks and learning. This is because poor smoothness can lead to poor performance bias.https://twitter.com/shakir_za/status/1337765774564483072 …
Still having fun at #NeurIPS2020
At 'I Can’t Believe It’s Not Better!' @MihaelaCRosca and a poster thinking about the fundamental role of smoothness in ML, and questions about definitions, measurements, complexity and learning paradigms. Your views? …https://i-cant-believe-its-not-better.github.io/ pic.twitter.com/Kp7rvTCHNl
At 'I Can’t Believe It’s Not Better!'