ICML 2019, Long Beach, June 12 th 2019 Session: Generative Models i - - PowerPoint PPT Presentation

▶
icml 2019 long beach june 12 th 2019 session generative
SMART_READER_LITE
LIVE PREVIEW

ICML 2019, Long Beach, June 12 th 2019 Session: Generative Models i - - PowerPoint PPT Presentation

Jun 03, 2023 191 likes •388 views

Luigi Antelmi 1 Nicholas Ayache 1 Philippe Robert 2,3 Marco Lorenzi 1 1 University of Cte d'Azur, Inria, Epione Project-Team, France. 2 University of Cte d'Azur, CoBTeK, France. 3 Centre Mmoire, CHU of Nice, France. Correspondence to:

slide-1
SLIDE 1 e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Luigi Antelmi1 Nicholas Ayache1 Philippe Robert2,3 Marco Lorenzi1

1University of Côte d'Azur, Inria, Epione Project-Team, France. 2University of Côte d'Azur, CoBTeK, France. 3Centre Mémoire, CHU of Nice, France.

Correspondence to: luigi.antelmi@inria.fr

ICML 2019, Long Beach, June 12th 2019 Session: Generative Models

slide-2
SLIDE 2

2

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Why I like Generative Models

“What I cannot create, I do not understand”

  • R. P. Feynman
slide-3
SLIDE 3

3

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

For C channels …

slide-4
SLIDE 4

4

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

For C channels we assume the following generative process: Decoders: reconstruction of data from the latent space z

slide-5
SLIDE 5

5

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

For C channels we assume the following generative process: Decoders: reconstruction of data from the latent space z Encoders: inference of the latent space z from the data

slide-6
SLIDE 6

6

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

For C channels we assume the following generative process: Decoders: reconstruction of data from the latent space z Encoders: inference of the latent space z from the data

slide-7
SLIDE 7

7

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

For C channels we assume the following generative process: Decoders: reconstruction of data from the latent space z Encoders: inference of the latent space z from the data

slide-8
SLIDE 8

8

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

Every channel is informative Decoders: reconstruction of data from the latent space z Encoders: inference of the latent space z from the data

slide-9
SLIDE 9

9

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

slide-10
SLIDE 10

10

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

Evidence Lower Bound

slide-11
SLIDE 11

11

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

Evidence Lower Bound

Encoding from a given channel

slide-12
SLIDE 12

12

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

Evidence Lower Bound

Encoding from a given channel Reconstruction of all the channels

slide-13
SLIDE 13

13

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

The Generative Multi-Channel Model

Evidence Lower Bound

Encoding from a given channel Reconstruction of all the channels Regularization inducing sparsity:

  • variational dropout on z
  • model selection
  • interpretability
  • pruning factor ~50%

Variational Dropout bibliography: Wang et al., ICML 2013; Kingma et al., NIPS 2015; Molchanov et al., ICML 2017.

slide-14
SLIDE 14

14

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Unsupervised clustering in Alzheimers’ Disease

Joint modeling of: Clinical scores + {Structural + Metabolic + Molecular} Imaging. Diagnosis status unknown to the model

Healthy Pathological Healthy Pathological

slide-15
SLIDE 15

15

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Generation from latent space

Adapted from: M. Lorenzi, Collège de France, 23/4/2019

z

Structural Imaging (MRI) Metabolic Imaging (FDG-PET) Molecular Imaging (AV45-PET)

  • Improved interpretability
  • Simulations for clinical trials
slide-16
SLIDE 16

16

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Generation from latent space

Adapted from: M. Lorenzi, Collège de France, 23/4/2019

z

Structural Imaging (MRI) Metabolic Imaging (FDG-PET) Molecular Imaging (AV45-PET)

  • Improved interpretability
  • Simulations for clinical trials
slide-17
SLIDE 17

17

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Generation from latent space

Adapted from: M. Lorenzi, Collège de France, 23/4/2019

z

Structural Imaging (MRI) Metabolic Imaging (FDG-PET) Molecular Imaging (AV45-PET)

  • Improved interpretability
  • Simulations for clinical trials
slide-18
SLIDE 18

18

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

Generation from latent space

Adapted from: M. Lorenzi, Collège de France, 23/4/2019

z

Structural Imaging (MRI) Metabolic Imaging (FDG-PET) Molecular Imaging (AV45-PET)

  • Improved interpretability
  • Simulations for clinical trials
slide-19
SLIDE 19

19

e-patient / e-medicine i n f
  • r
ma t i c s ma t h e m a t i c s

If you’re interested in: VAEs, Sparse Code, Interpretability, Prediction

  • f Missing Data, Medical Applications, …

see you at Poster #57, Pacific Ballroom 06:30 pm - ... luigi.antelmi@inria.fr Thank you!