SLIDE 1
DIVA: Domain Invariant Variational Autoencoders In collaboration - - PowerPoint PPT Presentation
DIVA: Domain Invariant Variational Autoencoders In collaboration - - PowerPoint PPT Presentation
DIVA: Domain Invariant Variational Autoencoders In collaboration with Jakub Tomczak, Christos Louizos and Max Welling Why do we care about domain generalization/invariance? Domain shift in medical imaging Patient 1 (Rajaraman et al., 2018)
SLIDE 2
SLIDE 3
Domain shift in medical imaging
(Rajaraman et al., 2018)
Patient 1
SLIDE 4
Domain shift in medical imaging
Patient 1 Malaria dataset
(Rajaraman et al., 2018)
SLIDE 5
Domain shift in medical imaging
Patient 1 Malaria dataset 1 cell == 1 image
(Rajaraman et al., 2018)
SLIDE 6
Domain shift in medical imaging
Patient 1 Malaria dataset 1 cell == 1 image Task: infected vs. uninfected
(Rajaraman et al., 2018)
SLIDE 7
Domain shift in medical imaging
Patient 1 Patient 2 Patient 3 Patient 4 Malaria dataset 1 cell == 1 image Task: infected vs. uninfected
(Rajaraman et al., 2018)
SLIDE 8
Can we disentangle the staining and the virus?
SLIDE 9
Disentanglement
(Kingma and Welling, 2014)
SLIDE 10
Disentanglement
(Kingma and Welling, 2014)
SLIDE 11
Disentanglement
(Kingma and Welling, 2014)
SLIDE 12
Disentanglement
(Kingma et al., 2014)
SLIDE 13
Disentanglement
(Kingma et al., 2014)
Two latents: z1 -> Content z2 -> Style
SLIDE 14
Disentanglement
(Kingma et al., 2014)
Two latents: z1 -> Content z2 -> Style Changing one doesn’t change the other
SLIDE 15
Disentanglement
(Kingma et al., 2014)
Two latents: z1 -> Content z2 -> Style Changing one doesn’t change the other Idea: Just use z1for classification
SLIDE 16
DIVA
SLIDE 17
DIVA
Generative Inference
SLIDE 18
DIVA
Generative Inference
Think of: d = patient, x = cell, y = infected/uninfected -> training tuple (x, y, d)
SLIDE 19
DIVA
Generative Inference
Think of: d = patient, x = cell, y = infected/uninfected -> training tuple (x, y, d)
SLIDE 20
Our model: DIVA
Generative Inference
Think of: d = patient, x = cell, y = infected/uninfected -> training tuple (x, y, d) Red: CNN for classification of y, dashed arrows == auxiliary classifiers
SLIDE 21
Our model: DIVA
Generative Inference
Think of: d = patient, x = cell, y = infected/uninfected -> training tuple (x, y, d) Red: CNN for classification of y, dashed arrows == auxiliary classifiers Green: Reconstruction of x
SLIDE 22
Our model: DIVA
Generative Inference
Think of: d = patient, x = cell, y = infected/uninfected -> training tuple (x, y, d) Red: CNN for classification of y, dashed arrows == auxiliary classifiers Green: Reconstruction of x Blue: Conditional prior distributions
SLIDE 23
Qualitative results
SLIDE 24
Qualitative results
SLIDE 25
Qualitative results
SLIDE 26
Qualitative results
SLIDE 27
Qualitative results
SLIDE 28
Quantitative results
SLIDE 29
Quantitative results
SLIDE 30
Unsupervised domains
SLIDE 31
Unsupervised domains
If I want to generalise to this patient
SLIDE 32
Unsupervised domains
If I want to generalise to this patient Does it help to have unlabeled data from this patient ?
SLIDE 33
Unsupervised domains
If I want to generalise to this patient Does it help to have unlabeled data from this patient ?
SLIDE 34