1 of 41

Variational Experimental Design Methods for Geophysical Applications

Dominik Strutz & Andrew Curtis

2 of 41

  • Can we design experiments to answer specific questions?
  • How do the data uncertainties affect optimal designs?

2

3 of 41

3

prior information

posterior information

-

information gain (data, design) =

What is the value of an experiment?

4 of 41

4

prior information

posterior information

-

information gain (data, design) =

5 of 41

5

prior information

posterior information

-

information gain (data, design) =

6 of 41

6

prior information

posterior information

-

information gain (data, design) =

7 of 41

7

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

8 of 41

8

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

9 of 41

9

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

10 of 41

10

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

11 of 41

11

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

12 of 41

12

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

13 of 41

13

prior information

posterior information

-

excepted information gain (design) =

average over all data plausible according to prior information

14 of 41

14

A quantity to assess the quality of an experimental design

prior information

posterior information

-

average over all data plausible according to prior information

excepted information gain (design) =

15 of 41

Variational Methods

15

16 of 41

Variational Methods

16

data

Mixture Density Network

17 of 41

17

  • Can we design experiments to answer specific questions?
  • How do the data uncertainties affect optimal designs?

18 of 41

18

What is the best design to constrain the earthquake hypocenter?

19 of 41

19

What is the best design to constrain the earthquake hypocenter?

20 of 41

20

What is the best design to constrain the earthquake hypocenter?

21 of 41

21

What is the best design to constrain the earthquake hypocenter?

22 of 41

22

What is the best design to constrain the earthquake hypocenter?

 

23 of 41

23

What is the best design to constrain the earthquake hypocenter?

 

24 of 41

24

What is the best design to constrain the earthquake depth?

 

 

25 of 41

25

What is the best design to constrain the earthquake depth?

 

What is the best design to constrain the earthquake epicenter?

26 of 41

26

What is the best design to constrain the earthquake epicenter?

 

27 of 41

27

What is the best design to constrain the earthquake epicenter?

 

28 of 41

28

  • Can we design experiments to answer specific questions?
  • How do the data uncertainties affect optimal designs?

29 of 41

Effect of the Likelihood Function

29

Lomax et al. (2014)

30 of 41

Effect of the Likelihood Function

30

Lomax et al. (2014)

31 of 41

Effect of the Likelihood Function

31

32 of 41

Effect of the Likelihood Function

32

damage zone

prior distribution of event locations

33 of 41

Effect of the Likelihood Function

33

34 of 41

Effect of the Likelihood Function

34

35 of 41

Effect of the Likelihood Function

35

36 of 41

Summary

  • Can we design experiments to answer specific questions?

  • How do the data uncertainties affect optimal designs?

Both questions of interest, and the form of data uncertainties, may significantly impact optimal designs

36

Visit my website (dominik-strutz.github.io) for further updates and contact information

37 of 41

37

38 of 41

38

39 of 41

AVO SGD example

39

40 of 41

AVO SGD example

40

41 of 41

AVO SGD example

41