Completed
When do larger models have high worst-group error?
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Neural Networks and Spurious Correlations
Automatically move to the next video in the Classroom when playback concludes
- 1 Intro
- 2 Typical machine learning pipeline
- 3 Taming the neural network beasts
- 4 In this talk...
- 5 Models latch onto spurious correlations
- 6 Improving worst-group error Simple and well-studied approach: reweighting loss based on group proportion
- 7 Talk outline
- 8 Empirical performance of reweighting Task: bird type Spurious background
- 9 When do larger models have high worst-group error?
- 10 Can larger models obtain low worst-group error?
- 11 Empirical performance of subsampling
- 12 Summary (spurious correlations)
- 13 Why do networks pick up on spurious correlations?
- 14 Simplicity bias of neural networks
- 15 How to overcome simplicity bias
- 16 Just train twice - results
- 17 Parting thoughts
