what do noisy datapoints tell us about the true signal? · 1/17/2012 · overview... . . ....
TRANSCRIPT
![Page 1: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/1.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What do noisy datapoints tell us about thetrue signal?
C. R. Hogg
November 16, 2011
1/31
![Page 2: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/2.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Outline
Overview
Bayesian Analysis
Gaussian Processes
Scattering Curves
Conclusions
2/31
![Page 3: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/3.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 4: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/4.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 5: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/5.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 6: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/6.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 7: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/7.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 8: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/8.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 9: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/9.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Overview
Q (1/A)
Sca
tte
red
In
ten
sity
0
500
1000
1500
2000
2500
2 3 4 5 6 7 8
Ernest Rutherford (1871-1937)
“If your experiment needs statistics,you ought to have done a better experiment”
(Or: use better statistics!)
3/31
![Page 10: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/10.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 11: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/11.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 12: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/12.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 13: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/13.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 14: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/14.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 15: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/15.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Goals for the talk
Uncertainty in single quantity:
Qu
an
tity
of
Inte
rest
7
8
9
10
11
12
13
. . . in continuous functions:
same shape
stays inside
1. Explain Bayesian analysis atconceptual level
2. Discuss quantifying uncertaintyin continuous functions
4/31
![Page 16: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/16.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Why Bayes at NIST?
NIST’s mission:To promote U.S. innovation and industrial
competitiveness by advancing measurementscience, standards, and technology in ways
that enhance economic security and improveour quality of life.
NIST’s vision:NIST will be the world’s leader in creating
critical measurement solutions and promotingequitable standards. Our efforts stimulate
innovation, foster industrial competitiveness,and improve the quality of life.
NIST’s core competencies:• Measurement science• Rigorous traceability• Development and use of standards
• Measurement isextremely important atNIST
• Must quantifyuncertainty:“A measurement result is
complete only when accompanied
by a quantitative statement of its
uncertainty.”a
• Which language todiscuss uncertainty?
• If “probabilities”:Bayesian analysis
aNIST TN 1297
5/31
![Page 17: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/17.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Why Bayes at NIST?
NIST’s mission:To promote U.S. innovation and industrial
competitiveness by advancing measurementscience, standards, and technology in ways
that enhance economic security and improveour quality of life.
NIST’s vision:NIST will be the world’s leader in creating
critical measurement solutions and promotingequitable standards. Our efforts stimulate
innovation, foster industrial competitiveness,and improve the quality of life.
NIST’s core competencies:• Measurement science• Rigorous traceability• Development and use of standards
• Measurement isextremely important atNIST
• Must quantifyuncertainty:“A measurement result is
complete only when accompanied
by a quantitative statement of its
uncertainty.”a
• Which language todiscuss uncertainty?
• If “probabilities”:Bayesian analysis
aNIST TN 1297
5/31
![Page 18: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/18.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Why Bayes at NIST?
NIST’s mission:To promote U.S. innovation and industrial
competitiveness by advancing measurementscience, standards, and technology in ways
that enhance economic security and improveour quality of life.
NIST’s vision:NIST will be the world’s leader in creating
critical measurement solutions and promotingequitable standards. Our efforts stimulate
innovation, foster industrial competitiveness,and improve the quality of life.
NIST’s core competencies:• Measurement science• Rigorous traceability• Development and use of standards
• Measurement isextremely important atNIST
• Must quantifyuncertainty:“A measurement result is
complete only when accompanied
by a quantitative statement of its
uncertainty.”a
• Which language todiscuss uncertainty?
• If “probabilities”:Bayesian analysis
aNIST TN 1297
5/31
![Page 19: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/19.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Why Bayes at NIST?
NIST’s mission:To promote U.S. innovation and industrial
competitiveness by advancing measurementscience, standards, and technology in ways
that enhance economic security and improveour quality of life.
NIST’s vision:NIST will be the world’s leader in creating
critical measurement solutions and promotingequitable standards. Our efforts stimulate
innovation, foster industrial competitiveness,and improve the quality of life.
NIST’s core competencies:• Measurement science• Rigorous traceability• Development and use of standards
• Measurement isextremely important atNIST
• Must quantifyuncertainty:“A measurement result is
complete only when accompanied
by a quantitative statement of its
uncertainty.”a
• Which language todiscuss uncertainty?
• If “probabilities”:Bayesian analysis
aNIST TN 12975/31
![Page 20: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/20.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?
“LIKELIHOOD”
2. How plausible is it?
“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?“POSTERIOR”
6/31
![Page 21: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/21.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?
“LIKELIHOOD”
2. How plausible is it?
“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?“POSTERIOR”
6/31
![Page 22: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/22.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?
“LIKELIHOOD”
2. How plausible is it?
“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?“POSTERIOR”
6/31
![Page 23: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/23.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?
“LIKELIHOOD”
2. How plausible is it?
“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?
“POSTERIOR”
6/31
![Page 24: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/24.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?“LIKELIHOOD”
2. How plausible is it?“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?“POSTERIOR”
6/31
![Page 25: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/25.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
What is Bayesian Analysis?
(Rev. Thomas Bayes, c. 1701 – 1761)
Essence of Bayes:• 2 questions for every
guess (i.e. every θ)1. How likely does it make
the actual data?“LIKELIHOOD”
2. How plausible is it?“PRIOR”
• Combine them to answerthe main question:
1. What is your newprobability, now thatyou’ve seen the data?“POSTERIOR”
6/31
![Page 26: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/26.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Likelihood of function p(y |f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Example: artificial dataset• Noise model: Poisson
p(y |f ) = f ye−f
y !
• Assume independentpixels
• Problem: not plausible• (What makes a function
“plausible”?)
7/31
![Page 27: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/27.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Likelihood of function p(y |f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Example: artificial dataset• Noise model: Poisson
p(y |f ) = f ye−f
y !
• Assume independentpixels
• Problem: not plausible• (What makes a function
“plausible”?)
7/31
![Page 28: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/28.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Likelihood of function p(y |f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Example: artificial dataset• Noise model: Poisson
p(y |f ) = f ye−f
y !
• Assume independentpixels
• Problem: not plausible• (What makes a function
“plausible”?)
7/31
![Page 29: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/29.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Likelihood of function p(y |f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Example: artificial dataset• Noise model: Poisson
p(y |f ) = f ye−f
y !
• Assume independentpixels
• Problem: not plausible• (What makes a function
“plausible”?)
7/31
![Page 30: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/30.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Likelihood of function p(y |f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Example: artificial dataset• Noise model: Poisson
p(y |f ) = f ye−f
y !
• Assume independentpixels
• Problem: not plausible• (What makes a function
“plausible”?)
7/31
![Page 31: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/31.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsity
(n
um
be
r o
f cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 32: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/32.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 33: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/33.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 34: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/34.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 35: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/35.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 36: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/36.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 37: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/37.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 38: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/38.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
“Plausibility” of function p(f )
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• Assume smooth andcontinuous
• No functional formassumed
• Naturally: unrelated to data
8/31
![Page 39: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/39.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 40: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/40.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 41: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/41.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 42: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/42.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 43: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/43.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 44: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/44.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)
Scattering vector Q (1/A)
Inte
nsi
ty (
nu
mb
er
of
cou
nts
)0
20
40
0.0 0.5 1.0
• “Best of both worlds”:a Plausible curves, whichb fit the data
• To represent uncertainty:show many guesses
• (Or, summarize them. . . )
9/31
![Page 45: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/45.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)Quantitative uncertainty visuals
0
5
10
15
20
25
30
35
40
0 0.2 0.4 0.6 0.8 1
Inte
nsi
ty (
no.
of
cou
nts
)
Q (1/A)
Noisy Data
9/31
![Page 46: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/46.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Posterior probability p(f |y)Quantitative uncertainty visuals
0
5
10
15
20
25
30
35
40
0 0.2 0.4 0.6 0.8 1
Inte
nsi
ty (
no.
of
cou
nts
)
Q (1/A)
Noisy DataTrue Curve!
9/31
![Page 47: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/47.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Recap: Bayesian denoisingPlausible curves
Scattering vector Q (1/A)
Inte
nsity
(num
ber
of co
unts
)0
20
40
0.0 0.5 1.0
10/31
![Page 48: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/48.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Recap: Bayesian denoisingCurves which fit the data
Scattering vector Q (1/A)
Inte
nsi
ty (
num
ber
of co
unts
)0
20
40
0.0 0.5 1.0
10/31
![Page 49: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/49.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Recap: Bayesian denoisingPlausible curves which fit the data
Scattering vector Q (1/A)
Inte
nsity
(num
ber
of co
unts
)0
20
40
0.0 0.5 1.0
10/31
![Page 50: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/50.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 51: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/51.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = ,
, , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 52: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/52.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , ,
, , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 53: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/53.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , ,
, ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 54: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/54.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , ,
,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 55: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/55.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 56: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/56.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
,
, , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 57: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/57.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, ,
, , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 58: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/58.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , ,
, . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 59: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/59.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 60: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/60.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 61: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/61.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 62: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/62.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 63: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/63.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 64: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/64.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 65: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/65.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 66: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/66.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 67: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/67.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Random variables; Random functions
?
p = 16
16
16
16
16
16
F = , , , , ,
, , , , . . .
x
F(x
)
• Random variable F :an uncertain quantity
• calculate probabilitiesfor its values
• take “random draws”(roll the die,flip the coin. . . )
• Random function F (x)?
11/31
![Page 68: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/68.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 69: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/69.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 70: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/70.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 71: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/71.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 72: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/72.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 73: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/73.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 74: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/74.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 75: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/75.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 76: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/76.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 77: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/77.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 78: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/78.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 79: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/79.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to think about “random functions”?
x
F(x
)
• Function: a collection ofindividual values
• Every value is a randomvariable, with. . .
1. variance2. correlation
• correlation × variance:covariance
• Gaussian Process:• Every point is a
Random Variable• Any (finite) subset has
Gaussian jointdistribution
12/31
![Page 80: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/80.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance Matrix
Str
ipe
In
ten
sity
= F
ea
ture
He
igh
t
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?
1. By individual entries2. As a whole
(central stripe)• Intensity:
height of features• Width:
width of features
13/31
![Page 81: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/81.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance Matrix
Str
ipe
In
ten
sity
= F
ea
ture
He
igh
t
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries
2. As a whole(central stripe)
• Intensity:height of features
• Width:width of features
13/31
![Page 82: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/82.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance Matrix
Str
ipe
In
ten
sity
= F
ea
ture
He
igh
t
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries
2. As a whole(central stripe)
• Intensity:height of features
• Width:width of features
13/31
![Page 83: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/83.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance Matrix
Str
ipe
In
ten
sity
= F
ea
ture
He
igh
t
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries
2. As a whole(central stripe)
• Intensity:height of features
• Width:width of features
13/31
![Page 84: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/84.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance Matrix
Str
ipe
In
ten
sity
= F
ea
ture
He
igh
t
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries2. As a whole
(central stripe)• Intensity:
height of features• Width:
width of features
13/31
![Page 85: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/85.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance MatrixS
trip
e I
nte
nsity
= F
ea
ture
He
igh
t
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries2. As a whole
(central stripe)• Intensity:
height of features• Width:
width of features
13/31
![Page 86: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/86.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
How to read a Covariance MatrixS
trip
e I
nte
nsity
= F
ea
ture
He
igh
t
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
• How to read the matrix?1. By individual entries2. As a whole
(central stripe)• Intensity:
height of features• Width:
width of features
13/31
![Page 87: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/87.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Example 1: Hydrocarbon combustion(Dave Sheen and Wing Tsang, NIST, Div. 632)
Hydrocarbon burningsimulations
• Need (many!) reaction rateconstants
• Measured individually• Predictions are precise,
quantitative
, wrong
14/31
![Page 88: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/88.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Example 1: Hydrocarbon combustion(Dave Sheen and Wing Tsang, NIST, Div. 632)
Hydrocarbon burningsimulations
• Need (many!) reaction rateconstants
• Measured individually• Predictions are precise,
quantitative, wrong
14/31
![Page 89: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/89.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Hydrocarbon combustion: flame speed experiments
Flame speed
Fuel-to-oxygen ratio
Speed (cm/s)
0
10
20
30
40
0.6 0.8 1.0 1.2 1.4 1.6
• Datapoints (fromseveral experiments)
• Model: lengthscales` and σf
• ±1σ range• See also: individual curves• But where did this model
come from. . . ?
15/31
![Page 90: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/90.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Hydrocarbon combustion: flame speed experiments
Flame speed
Fuel-to-oxygen ratio
Speed (cm/s)
0
10
20
30
40
0.6 0.8 1.0 1.2 1.4 1.6
• Datapoints (fromseveral experiments)
• Model: lengthscales` and σf
• ±1σ range• See also: individual curves• But where did this model
come from. . . ?
15/31
![Page 91: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/91.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Hydrocarbon combustion: flame speed experiments
Flame speed
Fuel-to-oxygen ratio
Speed (cm/s)
0
10
20
30
40
0.6 0.8 1.0 1.2 1.4 1.6
• Datapoints (fromseveral experiments)
• Model: lengthscales` and σf
• ±1σ range
• See also: individual curves• But where did this model
come from. . . ?
15/31
![Page 92: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/92.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Hydrocarbon combustion: flame speed experiments
Flame speed
Fuel-to-oxygen ratio
Speed (cm/s)
0
10
20
30
40
0.6 0.8 1.0 1.2 1.4 1.6
• Datapoints (fromseveral experiments)
• Model: lengthscales` and σf
• ±1σ range• See also: individual curves
• But where did this modelcome from. . . ?
15/31
![Page 93: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/93.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Hydrocarbon combustion: flame speed experiments
Flame speed
Fuel-to-oxygen ratio
Speed (cm/s)
0
10
20
30
40
0.6 0.8 1.0 1.2 1.4 1.6
• Datapoints (fromseveral experiments)
• Model: lengthscales` and σf
• ±1σ range• See also: individual curves• But where did this model
come from. . . ?
15/31
![Page 94: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/94.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 1:
flat
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor
• (slightly paraphrased inthe name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 95: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/95.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 1:
flat
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor• (slightly paraphrased in
the name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 96: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/96.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 1:
flat
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor• (slightly paraphrased in
the name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 97: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/97.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 1:
flat
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor• (slightly paraphrased in
the name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 98: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/98.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 2:
flat + line
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor• (slightly paraphrased in
the name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 99: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/99.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor, intro
MODEL 3:
flat + line + wiggle
• William of Occamc. 1288 - c. 1348
• Gave us Occam’s Razor• (slightly paraphrased in
the name of science)
• Claim: use probability,get this automatically
• (And, quantitative, too!)
• Example: 3 models. . .
16/31
![Page 100: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/100.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor in action
• Some models can explainmore datasets
• Each model is probabilitydistribution:
• Same total probabilityto distribute
• Which data actuallyobserved?
17/31
![Page 101: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/101.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor in action
• Some models can explainmore datasets
• Each model is probabilitydistribution:
• Same total probabilityto distribute
• Which data actuallyobserved?
17/31
![Page 102: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/102.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor in action
• Some models can explainmore datasets
• Each model is probabilitydistribution:
• Same total probabilityto distribute
• Which data actuallyobserved?
17/31
![Page 103: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/103.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor in action
• Some models can explainmore datasets
• Each model is probabilitydistribution:
• Same total probabilityto distribute
• Which data actuallyobserved?
17/31
![Page 104: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/104.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor in action
• Some models can explainmore datasets
• Each model is probabilitydistribution:
• Same total probabilityto distribute
• Which data actuallyobserved?
17/31
![Page 105: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/105.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 106: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/106.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 107: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/107.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 108: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/108.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 109: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/109.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 110: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/110.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 111: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/111.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 112: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/112.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 113: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/113.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 114: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/114.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Occam’s razor and Gaussian Processes
ℓℓℓ
σσ
σ
Probability
of Model:
0 1/9 1
• 9 models, varyingcomplexity
• Few datapoints (2):simple models preferred
• New data, some modelscan’t explain
• Three tiers1. Fit too poor2. Fit too good3. Just right
• Clear winner
18/31
![Page 115: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/115.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Example 2: Metal Strain
(Adam Creuziger and Mark Iadicola, NIST, Div. 655)
• Testing stress/strain ofsteels (auto parts, etc.)
• Clamp flat plate; pushupwards on middle
• Measure:1. Stress: X-ray diffraction2. Strain: Digital imaging
of spray-paint pattern
• Can’t painteverywhere!
(Figures courtesy of Mark Iadicola)
19/31
![Page 116: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/116.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Example 2: Metal Strain
(Adam Creuziger and Mark Iadicola, NIST, Div. 655)
• Testing stress/strain ofsteels (auto parts, etc.)
• Clamp flat plate; pushupwards on middle
• Measure:1. Stress: X-ray diffraction2. Strain: Digital imaging
of spray-paint pattern• Can’t paint
everywhere!
(Figures courtesy of Mark Iadicola)
19/31
![Page 117: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/117.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 118: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/118.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface
• Uncertainty bounds ±1σ• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 119: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/119.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 120: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/120.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 121: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/121.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 122: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/122.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 123: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/123.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 124: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/124.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Metal Strain: Preliminary Results
• Spheres representdatapoints
• Continuous surface• Uncertainty bounds ±1σ
• See also animations
• Competing model:anisotropic
• Occam’s razor lets uschoose!
• ∆ log(ML) = +183.4
• Suggestions forexperimental design
20/31
![Page 125: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/125.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Need to extend the modelS
trip
e I
nte
nsity
= F
ea
ture
He
igh
t
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
• Recall: how to readcovariance matrices“as a whole”
• Intensity:height of features
• Width:width of features
• Not flexible enough forreal data!
21/31
![Page 126: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/126.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Need to extend the modelS
trip
e I
nte
nsity
= F
ea
ture
He
igh
t
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Str
ipe
Wid
th =
Fe
atu
re W
idth
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
• Recall: how to readcovariance matrices“as a whole”
• Intensity:height of features
• Width:width of features
• Not flexible enough forreal data!
21/31
![Page 127: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/127.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Two extensions
Va
ryin
g W
idth
s
X
-100
010
0
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Mu
ltip
le C
on
trib
utio
ns
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• Two main extensions. . .
1. Varying Feature widths• ` → `(X )
2. Multiple contributions• Background everywhere• Localized “peak” regions
22/31
![Page 128: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/128.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Two extensionsV
ary
ing
Wid
ths
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Mu
ltip
le C
on
trib
utio
ns
X
-100
01
00
0 1 2 3X
01
23
0 1 2 3
• Two main extensions. . .1. Varying Feature widths
• ` → `(X )
2. Multiple contributions• Background everywhere• Localized “peak” regions
22/31
![Page 129: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/129.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Two extensionsV
ary
ing
Wid
ths
X-1
00
0100
0 1 2 3X
01
23
0 1 2 3
Cov
2500
0
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
Mu
ltip
le C
on
trib
utio
ns
X
-100
0100
0 1 2 3X
01
23
0 1 2 3
• Two main extensions. . .1. Varying Feature widths
• ` → `(X )
2. Multiple contributions• Background everywhere• Localized “peak” regions
22/31
![Page 130: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/130.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Simulated XRD from 2 nm Au nanoparticles
0
500
1000
1500
2000
2500
0 5 10 15 20 25 30 35
I(Q
)
True function
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
True function
• Core/shell structure• Shell atoms vibrate more• Correlated thermal
motion(Signature: hi-Qoscillations)
• Problem: Poisson noiseswamps these oscillations!
• Changing feature widths:use `(Q)
X
01
23
0 1 2 3
23/31
![Page 131: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/131.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Simulated XRD from 2 nm Au nanoparticles
0
500
1000
1500
2000
2500
0 5 10 15 20 25 30 35
I(Q
)
True function
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
True function
• Core/shell structure• Shell atoms vibrate more• Correlated thermal
motion(Signature: hi-Qoscillations)
• Problem: Poisson noiseswamps these oscillations!
• Changing feature widths:use `(Q)
X
01
23
0 1 2 3
23/31
![Page 132: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/132.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Simulated XRD from 2 nm Au nanoparticles
0
500
1000
1500
2000
2500
0 5 10 15 20 25 30 35
I(Q
)
True function
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
True function
• Core/shell structure• Shell atoms vibrate more• Correlated thermal
motion(Signature: hi-Qoscillations)
• Problem: Poisson noiseswamps these oscillations!
• Changing feature widths:use `(Q)
X
01
23
0 1 2 3
23/31
![Page 133: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/133.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Simulated XRD from 2 nm Au nanoparticles
0
500
1000
1500
2000
2500
0 5 10 15 20 25 30 35
I(Q
)
Noisy dataTrue function
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataTrue function
• Core/shell structure• Shell atoms vibrate more• Correlated thermal
motion(Signature: hi-Qoscillations)
• Problem: Poisson noiseswamps these oscillations!
• Changing feature widths:use `(Q)
X
01
23
0 1 2 3
23/31
![Page 134: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/134.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Simulated XRD from 2 nm Au nanoparticles
0
500
1000
1500
2000
2500
0 5 10 15 20 25 30 35
I(Q
)
Noisy dataTrue function
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataTrue function
• Core/shell structure• Shell atoms vibrate more• Correlated thermal
motion(Signature: hi-Qoscillations)
• Problem: Poisson noiseswamps these oscillations!
• Changing feature widths:use `(Q)
X
01
23
0 1 2 3
23/31
![Page 135: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/135.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Denoising results
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataTrue function
AWS (raw data)• AWS: jagged; loses signal
at Q = 26A−1
• Wavelets: smooth, but stilllose signal
• Bayes: also smooth, butkeeps signal
• Uncertainty boundscapture true function
24/31
![Page 136: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/136.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Denoising results
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataTrue function
Wavelets• AWS: jagged; loses signal
at Q = 26A−1
• Wavelets: smooth, but stilllose signal
• Bayes: also smooth, butkeeps signal
• Uncertainty boundscapture true function
24/31
![Page 137: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/137.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Denoising results
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataTrue function
Bayesian• AWS: jagged; loses signal
at Q = 26A−1
• Wavelets: smooth, but stilllose signal
• Bayes: also smooth, butkeeps signal
• Uncertainty boundscapture true function
24/31
![Page 138: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/138.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Denoising results
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataBayesian
True functionAWS (raw data)
Wavelets
• AWS: jagged; loses signalat Q = 26A−1
• Wavelets: smooth, but stilllose signal
• Bayes: also smooth, butkeeps signal
• Uncertainty boundscapture true function
24/31
![Page 139: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/139.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Residuals
Residuals vs. noisy data
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesi
an m
ean
Wa
vele
ts
Bayesi
an d
raw
s
0.240 0.245 0.250
Residuals vs. true curve
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesi
an m
ean
Wa
vele
ts
Bayesi
an d
raw
s
Tru
e c
urv
e
0.000 0.005 0.010
• Need: global fidelitymeasure
• Mean square residuals . . .1. vs. noisy data
• AWS looks best2. vs. true curve
• Bayes is best• AWS “good” score:
was overfitting noise!
25/31
![Page 140: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/140.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Residuals
Residuals vs. noisy data
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesi
an m
ean
Wa
vele
ts
Bayesi
an d
raw
s0.240 0.245 0.250
Residuals vs. true curve
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesi
an m
ean
Wa
vele
ts
Bayesi
an d
raw
s
Tru
e c
urv
e
0.000 0.005 0.010
• Need: global fidelitymeasure
• Mean square residuals . . .1. vs. noisy data
• AWS looks best
2. vs. true curve• Bayes is best• AWS “good” score:
was overfitting noise!
25/31
![Page 141: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/141.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Residuals
Residuals vs. noisy data
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesi
an m
ean
Wa
vele
ts
Bayesi
an d
raw
s0.240 0.245 0.250
Residuals vs. true curve
MSR
Pro
ba
bili
ty D
en
sity
0
AW
S
AW
S (
raw
data
)
Bayesia
n m
ean
Wa
vele
ts
Bayesi
an d
raw
s
Tru
e c
urv
e
0.000 0.005 0.010
• Need: global fidelitymeasure
• Mean square residuals . . .1. vs. noisy data
• AWS looks best2. vs. true curve
• Bayes is best• AWS “good” score:
was overfitting noise!
25/31
![Page 142: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/142.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
TiO2 nanoparticles
NIST SRM 1898:
(Ohno et al., J. Catalysis, 2011)
• X-ray powder diffractionfrom 20 nm TiO2nanoparticles
• Motivations:1. Real-world example
(Violates ourassumptions)
2. More difficult data(contains feature-freebackground regions)
26/31
![Page 143: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/143.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Comparing the fits
Q (1/A)
0
1000
2000
3000
1 2 3 4 5 6 7
Q (1/A)
200
400
600
800
1000
3.2 3.3 3.4
• All handle sharp peaks• Every technique misses a
few features: AWS,wavelets, even Bayes
27/31
![Page 144: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/144.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Comparing the fits
Q (1/A)
0
1000
2000
3000
1 2 3 4 5 6 7
Q (1/A)
80
100
120
5.8 5.9
• All handle sharp peaks• Every technique misses a
few features: AWS,wavelets, even Bayes
27/31
![Page 145: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/145.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Comparing the fits
Q (1/A)
0
1000
2000
3000
1 2 3 4 5 6 7
Q (1/A)
100
200
7.0 7.2
• All handle sharp peaks• Every technique misses a
few features: AWS,wavelets, even Bayes
27/31
![Page 146: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/146.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Comparing the fits
Q (1/A)
0
1000
2000
3000
1 2 3 4 5 6 7
Q (1/A)
100
120
6.2 6.4
• All handle sharp peaks• Every technique misses a
few features: AWS,wavelets, even Bayes
27/31
![Page 147: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/147.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Comparing the fits
Q (1/A)
0
1000
2000
3000
1 2 3 4 5 6 7
Q (1/A)
80
100
120
5.2 5.3
• All handle sharp peaks• Every technique misses a
few features: AWS,wavelets, even Bayes
27/31
![Page 148: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/148.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
ResidualsResiduals vs. quasi-true curve
MSR
Pro
ba
bili
ty D
en
sity
AW
S
Bayesi
an M
ean
Wa
vele
ts
0.025 0.030 0.035
Training on
Even Points
Mean MSR
Ove
rfitt
ing
0.0
00
0.0
20
Baye
sA
WS
Wa
vele
ts
0.170 0.180
Training on
Odd Points
Mean MSR
Ove
rfitt
ing
0.0
00
0.0
20
Baye
sA
WS
Wa
vele
ts
0.170 0.180
• Bayes single-curvecomparable to benchmarks
• Cross-validation:(Checking for overfitting)Bayes is best. . .
1. In both categories2. For both training sets
28/31
![Page 149: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/149.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
ResidualsResiduals vs. quasi-true curve
MSR
Pro
ba
bili
ty D
en
sity
AW
S
Bayesi
an M
ean
Wa
vele
ts
0.025 0.030 0.035
Training on
Even Points
Mean MSR
Ove
rfittin
g0.0
00
0.0
20
Baye
sA
WS
Wa
vele
ts
0.170 0.180
Training on
Odd Points
Mean MSR
Ove
rfittin
g0.0
00
0.0
20
Bayes
AW
S
Wa
vele
ts
0.170 0.180
• Bayes single-curvecomparable to benchmarks
• Cross-validation:(Checking for overfitting)Bayes is best. . .
1. In both categories2. For both training sets
28/31
![Page 150: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/150.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Recap: Bayesian Concepts
Scattering vector Q (1/A)
Inte
nsity
(n
um
be
r o
f co
un
ts)
02
04
0
0.0 0.5 1.0
• Bayesian analysis:using probabilitiesto describe uncertainty
• choose answers withboth plausibilityand data fit
• a natural framework formodel selectionconcepts(Occam’s razor)
29/31
![Page 151: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/151.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Recap: Uncertainty in continuous functions
110
120
130
140
150
160
170
23 24 25 26 27 28 29 30
Q (1/A)
Noisy dataBayesian
True functionAWS (raw data)
Wavelets
• Gaussian Processes: canstipulate smoothness,without worrying aboutfunctional form
• Open-source softwarepackage
• Very flexible: can help avariety of projects
30/31
![Page 152: What do noisy datapoints tell us about the true signal? · 1/17/2012 · Overview... . . . Bayesian Analysis. . . . .. . . . . Gaussian Processes. .. . .. . . Scattering Curves](https://reader036.vdocuments.us/reader036/viewer/2022081613/5fb9ba5e905ec326a17950d8/html5/thumbnails/152.jpg)
. .Overview
.
.
. . . .
Bayesian Analysis. . . . .. . . . .
Gaussian Processes. .. . .. . .
Scattering Curves. . .Conclusions
Acknowledgements
• Team members: Igor Levin, Kate Mullen• Collaborators:
• Flame speed : Dave Sheen, Wing Tsang• Metal Strain: Adam Creuziger, Mark Iadicola
• WERB readers: Victor Krayzmann, Adam Pintar• Statistical guidance: Antonio Possolo, Blaza Toman
31/31