Underspecification arises when a model is validated on a held-out data that doesn’t reflect what it will face in deployment, which can lead to poor performance in new settings. Learn about the causes of this effect and potential mitigation strategies ↓ https://t.co/Yl6tyoHyH6

Favorite tweet:

Underspecification arises when a model is validated on a held-out data that doesn’t reflect what it will face in deployment, which can lead to poor performance in new settings. Learn about the causes of this effect and potential mitigation strategies ↓ https://t.co/Yl6tyoHyH6

— Google AI (@GoogleAI) Oct 18, 2021

Underspecification arises when a model is validated on a held-out data that doesn’t reflect what it will face in deployment, which can lead to poor performance in new settings. Learn about the causes of this effect and potential mitigation strategies ↓ https://t.co/Yl6tyoHyH6 https://twitter.com/GoogleAI/status/1450210565796806662 https://t.co/Yl6tyoHyH6 October 19, 2021 at 06:21AM