Completed
Black Box Shift Correction (CIFAR10 w IW-ERM)
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Robust Deep Learning Under Distribution Shift
Automatically move to the next video in the Classroom when playback concludes
- 1 Intro
- 2 Outline
- 3 Standard assumptions
- 4 Adversarial Misspellings (Char-Level Attack)
- 5 Curated Training Task Fail to Represent Reality
- 6 Feedback Loops
- 7 Impossibility absent assumptions
- 8 Detecting and correcting for label shift with black box predictors
- 9 Motivation 1: Pneumonia prediction
- 10 Epidemic
- 11 Motivation 2: Image Classification
- 12 The test-Item effect
- 13 Domain Adaptation - Formal Setup
- 14 Label Shift (aka Target Shift)
- 15 Contrast with Covariate Shift
- 16 Black Box Shift Estimation (BBSE)
- 17 Confusion matrices
- 18 Applying the label shift assumption...
- 19 Consistency
- 20 Error bound
- 21 Detection
- 22 Estimation error (MNIST)
- 23 Black Box Shift Correction (CIFAR10 w IW-ERM)
- 24 A General Pipeline for Detecting Shift
- 25 Non-adversarial image perturbations
- 26 Detecting adversarial examples
- 27 Covariate shift + model misspecification
- 28 Implicit bias of SGD on linear networks w. linearly separable data
- 29 Impact of IW on ERM decays over MLP training
- 30 Weight-Invariance after 1000 epochs
- 31 L2 Regularization v Dropout
- 32 Deep DA / Domain-Adversarial Nets
- 33 Synthetic experiments
