1 of 16

MADNESS

Maximum-A-posteriori solution with Deep generative NEtworks

for Source Separation

Biswajit Biswas, Eric Aubourg, Alexandre Boucaud, Axel Guinot, Junpeng Lao, Cécile Roucelle

BDL Workshop

23rd June, 2022

2 of 16

Surveys and Challenges

Large survey of Space and Time (LSST) at Vera Rubin Observatory:

  • Ground-based
  • constrain Dark Energy
  • 3.2 billion pixel camera
  • 6 observation bands in visible range

more depth + area of coverage ⇒ More statistics!

greater depth means more complex data!

~ Galaxies (60% in LSST ) are expected to overlap (blending) in images due to increased depth

HSC ultradeep image

Page 2

3 of 16

ML for Deblending

Arcelin et al (arXiv:2005.12039)

Page 2

4 of 16

Denoising (Single source)

Where, 𝑥* is the maximum a posteriori probability (MAP) estimate

Input image (𝚢)

Predicted image (𝑥)

Residual (𝚢-𝑥)

Page 3

5 of 16

Train VAE as generative model

𝑥

𝑥

For example: Lanusse et al (arXiv:2008.03833)

Page 4

Training:

Regularization term

Reconstruction term

6 of 16

MAP estimate in latent space

Where, 𝑧* is the maximum a posteriori probability estimate in the latent space

Going to the latent space

Predicted galaxy ( x )

Input image ( 𝚢 )

Page 5

7 of 16

How to choose a prior?

Page 6

8 of 16

Normalizing flow

Page 7

9 of 16

Minimization

  • Start with random z
  • Do gradient descent in the latent space to minimize the objective function

Predicted galaxy ( x )

Input image ( 𝚢 )

Where, 𝑧* is the maximum a posteriori probability estimate in the latent space

Page 8

10 of 16

Deblending (Multiple sources)

Reconstructed field

Probability that predictions are galaxies!

Page 9

11 of 16

Deblending Example

Input image

Predictions

Residual image

(input - predictions)

Page 10

12 of 16

Moving to a larger field…

Input field (501 x 501)

Sinh-1 ( Input field )

Page 11

13 of 16

Input field

predicted field

residual field

(Input - predicted)

Moving to a larger field…

Page 12

Preliminary results*

14 of 16

Compare with SOTA

Page 13

15 of 16

Conclusion and Future work

  • Developed MAP estimate with generative models

  • Performance close to SOTA!

  • Benchmark speed

  • metrics

  • Real data!

Page 14

16 of 16

Thank you!