Episode 20: Hattie Zhou, Mila, on supermasks, iterative learning, and fortuitous forgetting
Generally Intelligent
Getting Rid of Specifics for Spurious Features
It's really hard to define what spurious features are in the general case or even if they're not useful because I think humans still use spurious features. But yeah, getting rid of specifics for spurious features could be very valuable, very doable. Like, for example, things with fairness, right? You don't want the model to use rays, gender and features like that.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
