Completed
For classification problems, "natural distribution shifts are an open research problem"
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
The Role of Data and Models for Deep-Learning Based Image Reconstruction
Automatically move to the next video in the Classroom when playback concludes
- 1 Intro
- 2 Scaling language models
- 3 Data set sizes in imaging tasks are small
- 4 Expected performance behavior
- 5 U-net-based denoising
- 6 U-net-based accelerated MRI
- 7 Swin transformer based denoising
- 8 Reconstruction methods
- 9 What we might expect
- 10 Dataset shift
- 11 Adversarially filtered shift
- 12 Goal: improve performance unter distribution shifts
- 13 For classification problems, "natural distribution shifts are an open research problem"
- 14 Improving performance for 11-minimization is easy
- 15 Test time training
- 16 Closing the distribution shift performance gap for anatomy shift
- 17 Closing the distribution shift performance gap with test-time-training
- 18 References