The Robot Brains Podcast Chelsea Finn: how to build AI that can keep up with an always changing world
29 snips
Mar 22, 2023 AI Snips
Chapters
Transcript
Episode notes
Distribution Shift Challenge
- AI struggles with distribution shift, where real-world data differs from training data.
- This hinders deployment in changing environments, like spam filters adapting to evolving spam.
Robot Cup Example
- A robot trained to pick up a cup in a lab may fail with a different cup or environment.
- This illustrates AI's current limitations in generalizing to new situations due to distribution shift.
Improving AI Robustness
- Two approaches to improving AI robustness: focusing on spurious features and enabling on-the-fly adaptation.
- Spurious features, like backgrounds in image classification, can mislead models, while adaptation allows for continuous learning.