SlideShare a Scribd company logo
1 of 28
Download to read offline
1/28
Robust Joint and Individual Component Analysis
Alina Leidinger
Imperial College London
June 8, 2016
Alina Leidinger (ICL) RJICA June 8, 2016 1 / 28
2/28
Joint and Individual Variation Explained (JIVE)
1 2 3 4
1
http://www.uni-regensburg.de/Fakultaeten/phil_Fak_II/Psychologie/
Psy_II/beautycheck/english/durchschnittsgesichter/m(01-32)_gr.jpg.
2
http:
//img13.deviantart.net/47f8/i/2014/177/5/a/sketch___male_face_by_pmucks-
d7o53mf.png.
3
https://www.etc.cmu.edu/projects/atl/images/facial/plainface.jpg.
4
http://www.thermology.com/infrared22.JPG.
Alina Leidinger (ICL) RJICA June 8, 2016 2 / 28
3/28
Contents
Joint and Individual Variation Explained (JIVE)
Robust JIVE
Robust Nuclear Norm Regularised JIVE
Toy Example
Alina Leidinger (ICL) RJICA June 8, 2016 3 / 28
4/28
Joint and Individual Variation Explained (JIVE)5
different views on the same data e.g. measurements of different
quantities on the same set of objects
Original application: cancer research
Data from The Cancer Genome Atlas
Potential fields of Application: clinical studies (patient groups),
Finance (markets), e-commerce (websites), environmental sciences
(locations), facial recognition (expressions) ...
5
Eric F. Lock et al. “Joint and individual variation explained (JIVE) for integrated
analysis of multiple data types”. In: Annals of Applied Statistics 7.1 (2013). ID:
TN_scopus2-s2.0-84876058478, pp. 523–542.
Alina Leidinger (ICL) RJICA June 8, 2016 4 / 28
5/28
JIVE
suitability for high dimensional data n < pi
Alina Leidinger (ICL) RJICA June 8, 2016 5 / 28
6/28
JIVE: Mathematical Framework
Xi = Ji + Ai + Ei i = 1, . . . , k
or



X1
...
Xk



X
=



J1
...
Jk



J
+



A1
...
Ak



A
+



E1
...
Ek



E
require uncorrelatedness between joint and individual components:
JAT
i = 0
Alina Leidinger (ICL) RJICA June 8, 2016 6 / 28
7/28
JIVE: Optimisation Problem
arg min
J,Ai
X − J − [AT
1 , . . . , AT
k ]T 2
F
s.t. JAT
i = 0, rank(J) = r, rank(Ai ) = ri
J(t+1)
= arg min
J
J − (X − [A
(t)T
1 , . . . , A
(t)T
k ]T
) 2
F
s.t. rank(J) = r
A
(t+1)
i = arg min
Ai
Ai − (Xi − J
(t+1)
i ) 2
F
s.t. rank(Ai ) = ri , J(t+1)
AT
i = 0
Alina Leidinger (ICL) RJICA June 8, 2016 7 / 28
8/28
JIVE: Alternating Optimisation
J(t+1)
= arg min
J
J − (X − [A
(t)T
1 , . . . , A
(t)T
k ]T
) 2
F s.t. rank(J) = r
→ rank-r SVD of X − [A
(t)T
1 , . . . , A
(t)T
k ]T
A
(t+1)
i = arg min
Ai
Ai − (Xi − J
(t+1)
i ) 2
F s.t rank(Ai ) = ri , J(t+1)
AT
i = 0
→ (rank-ri SVD of Xi − J
(t+1)
i ) × (I − VV T )
Singular Value Decomposition: J(t+1) = UΣV T
Alina Leidinger (ICL) RJICA June 8, 2016 8 / 28
9/28
JIVE: Algorithm6
Algorithm 1 JIVE
1: Xjoint = X = [XT
1 . . . XT
k ]T
2: while not converged do
3: Estimate J by a rank r SVD of Xjoint i.e. J = Ur Λr V T
r
4: for i = 1, . . . , k do
5: Set XIndividual
i = Xi − Ji
6: Estimate Ai by a rank ri SVD of XIndividual
i (I − VV T )
7: Set XJoint
i = Xi − Ai
8: XJoint = [XJointT
1 . . . XJointT
k ]T
6
Lock et al. “Supplement to "JOINT AND INDIVIDUAL VARIATION EXPLAINED
(JIVE) FOR INTEGRATAED ANALYSIS OF MULTIPLE DATA TYPES"”. In: The
Annals of Applied Statistics 7.1 (2013).
Alina Leidinger (ICL) RJICA June 8, 2016 9 / 28
10/28
JIVE: Criticism
sensitive to outliers, missing data points
sensitive to gross non-Gaussian noise
combinatorial pre-processing step: inefficient, unsuitable for analysis of
images
Alina Leidinger (ICL) RJICA June 8, 2016 10 / 28
11/28
Robust JIVE
arg min
J,Ai
E 0 s.t. X = J + A + E, JAT
i = 0
→ arg min
J,Ai
E 1 s.t. X = J + A + E, JAT
i = 0
non-Gaussian noise
sparsity
0-norm → 1-norm7
7
Robert Tibshirani. “Regression Shrinkage and Selection via the Lasso”. In: Journal
of the Royal Statistical Society.Series B (Methodological) 58.1 (1996). ID:
TN_jstor_archive2346178, pp. 267–288.
Alina Leidinger (ICL) RJICA June 8, 2016 11 / 28
12/28
Robust JIVE: Augmented Lagrangian Method8
arg max
X
f (X) s.t. g(X) = 0
X(t)
= arg max
X
f (X) + Z(t)
, g(X) +
µ(t)
2
g(X) 2
F
= arg max
X
f (X) +
µ(t)
2
g(X) + Z(t)
/µ(t) 2
F
µ(t+1)
= ρµ(t)
Z(t+1)
= Z(t)
+ µ(t)
(g(X))
8
Dimitri P. Bertsekas. Constrained optimization and Lagrange multiplier methods.
Bibliography: p.383-392.- Includes index.; ID: 44IMP_ALMA_DS2145810210001591.
New York ; London: Academic Press, 1982.
Alina Leidinger (ICL) RJICA June 8, 2016 12 / 28
13/28
Robust JIVE: Alternating Direction Method9
arg min
E
E 1+
µ
2
X − J − A − E + Z/µ 2
F
s.t. JAT
i = 0, rank(J) = r, rank(Ai ) = ri
E∗
= arg min
E
E 1+
µ
2
X − J − A − E + Z/µ 2
F
J∗
= arg min
J
µ
2
X − J − A − E + Z/µ 2
F s.t. rank(J) = r
A∗
i = arg min
Ai
µ
2
Xi − Ji − Ai − Ei + Zi /µ 2
F
s.t rank(Ai ) = ri , JAT
i = 0
9
Daniel Gabay and Bertrand Mercier. “A dual algorithm for the solution of nonlinear
variational problems via finite element approximation”. In: Computers and Mathematics
with Applications 2.1 (1976). ID:
TN_sciversesciencedirect_elsevier0898-1221(76)90003-1, pp. 17–40.
Alina Leidinger (ICL) RJICA June 8, 2016 13 / 28
14/28
Robust JIVE: ADM Steps10
E(t+1)
= arg min
E
E 1+
µ(t)
2
X − J(t)
− A(t)
− E + Z(t)
/µ(t) 2
F
= Sµ(t)−1 (X − J(t)
− A(t)
+ Z(t)
/µ(t)
)
Sa(x) =



x − a x > a
x + a x < a
0 otherwise
10
Emmanuel Candès J. et al. “Robust principal component analysis?” In: Journal of
the ACM (JACM) 58.3 (2011). ID: TN_acm1970395, pp. 1–37.
Alina Leidinger (ICL) RJICA June 8, 2016 14 / 28
15/28
Robust JIVE: Algorithm
Algorithm 2 Robust JIVE
1: Output: J, Ai , E
2: Xjoint = X = [XT
1 . . . XT
k ]T
3: Initialise µ(0) = 0.01, Z(0) = 0, E(0) = 0, J(0) = 0, A
(0)
i = 0, ρ = 1.1
4: while not converged do
5: Set E(t+1) = Sµ(t)−1 (X − J(t) − A(t) + Z(t)/µ(t))
6: Set J(t+1) equal to a rank r SVD of X − A(t) − E(t+1) + Z(t)/µ(t)
7: for i = 1, . . . , k do
8: Set A
(t+1)
i equal to a rank ri SVD of (Xi − J
(t+1)
i − E
(t+1)
i +
Z
(t)
i /µ(t))(I − VV T ) where J(t+1) = UΣV T
9: Update Z(t+1) = Z(t) + µ(t)(X − J(t+1) − A(t+1) − E(t+1))
10: Update µ(t+1) = µ(t)ρ
11: t = t + 1
Alina Leidinger (ICL) RJICA June 8, 2016 15 / 28
16/28
Robust Nuclear Norm Regularised JIVE
minJ,Ai ,E αrank(J) + i αi rank(Ai ) + λ E 1
s.t. X = J + A + E, JAT
i = 0
rank function → nuclear norm, . ∗
11
11
Benjamin Recht, Maryam Fazel, and Pablo A. Parrilo. “Guaranteed Minimum- Rank
Solutions of Linear Matrix Equations via Nuclear Norm Minimization”. In: SIAM
Review 52.3 (2010). ID: TN_siam10.1137/070697835, pp. 471–501.
Alina Leidinger (ICL) RJICA June 8, 2016 16 / 28
17/28
Robust Nuclear Norm Regularised JIVE
min
J,Ai ,E
α J ∗+
i
αi Ai ∗+λ E 1
s.t. X = J + A + E, JAT
i = 0
or min
J,Ai ,Ri ,E
α J ∗+
i
αi Ri ∗+λ E 1
s.t. X = J + A + E, Ai = Ri , JAT
i = 0
= min
J,Ai ,Ri ,E
α J ∗+
i
αi Ri ∗+λ E 1
+
µ
2
X − J − A − E +
Z
µ
2
F +
i
γi
2
Ri − Ai +
Yi
γi
2
F s.t.JAT
i = 0
Alina Leidinger (ICL) RJICA June 8, 2016 17 / 28
18/28
Robust Nuclear Norm Regularised JIVE: ADM Steps12
E(t+1)
= arg min
E
λ E 1+
µ(t)
2
E − (X − J(t+1)
− A(t+1)
+
Z(t)
µ(t)
) 2
F
= Sλ/µ(t) (X − J(t+1)
− A(t+1)
+ Z(t)
/µ(t)
)
12
Emmanuel Candès J. et al. “Robust principal component analysis?” In: Journal of
the ACM (JACM) 58.3 (2011). ID: TN_acm1970395, pp. 1–37.
Alina Leidinger (ICL) RJICA June 8, 2016 18 / 28
19/28
Robust Nuclear Norm Regularised JIVE13
J(t+1)
= arg min
J
α J ∗+
µ(t)
2
J − (X − A(t)
− E(t)
+
Z(t)
µ(t)
) 2
F
= Dα/µ(t) (X − A(t)
− E(t)
+
Z(t)
µ(t)
)
R
(t+1)
i = arg min
Ri
αi Ri ∗+
γ
(t)
i
2
Ri − A
(t)
i +
Y
(t)
i
γ
(t)
i
2
F
= Dαi /γ
(t)
i
(A
(t+1)
i −
Y
(t)
i
γ
(t)
i
)
Dλ(X) = Udiag(max{σi − λ, 0})V T
(X = UΣV T
)
13
Jian-Feng Cai, Emmanuel J. Candès, and Zuowei Shen. “A Singular Value
Thresholding Algorithm for Matrix Completion”. In: SIAM Journal on Optimization
20.4 (2010). ID: TN_siam10.1137/080738970, pp. 1956–1982.
Alina Leidinger (ICL) RJICA June 8, 2016 19 / 28
20/28
Robust Nuclear Norm Regularised JIVE:
Linearising Cost Function1415
arg min
Ai
µ
2
Xi − Ji − Ai − Ei +
Zi
µ
2
F +
γi
2
Ri − Ai +
Yi
γi
2
F
f (Ai )
s.t.JAT
i = 0
f (Ai ) ≈ f (A
(t)
i ) + f (A
(t)
i ), Ai − A
(t)
i +
1
2τ
Ai − A
(t)
i
2
F
= f (A
(t)
i ) +
1
2τ
Ai − A
(t)
i + τ f (A
(t)
i ) 2
F
14
Junfeng Yang and Xiaoming Yuan. “Linearized augmented Lagrangian and
alternating direction methods for nuclear norm minimization”. In: Mathematics of
Computation 82.281 (2013), pp. 301–329.
15
Zhouchen Lin, Risheng Liu, and Zhixun Su. “Linearized Alternating Direction
Method with Adaptive Penalty for Low-Rank Representation”. In: (2011). ID:
TN_arxiv1109.0367.
Alina Leidinger (ICL) RJICA June 8, 2016 20 / 28
21/28
Robust Nuclear Norm Regularised JIVE:
Linearising Cost Function
A
(t+1)
i = arg min
Ai
1
2τ
Ai − A
(t)
i + τ f (A
(t)
i ) 2
F s.t. JAT
i = 0
= (A
(t)
i − τ f (A
(t)
i ))(I − VV T
) J(t+1)
= UΣV T
f (A
(t)
i ) =µ(t)
(A
(t)
i − Xi + J
(t+1)
i − Z
(t)
i /µ(t)
)
+ γ
(t)
i (A
(t)
i − R
(t)
i − Y
(t)
i /γ
(t)
i ))
Alina Leidinger (ICL) RJICA June 8, 2016 21 / 28
22/28
Toy Example: Artworks
16 Alina Leidinger (ICL) RJICA June 8, 2016 22 / 28
23/28
Toy Example
Figure: Superposed Images
17
17
Joint and Individual Variation Explained (JIVE) for the Integrated Analysis of
Multiple Datatypes. http://www.tc.umn.edu/~elock/Talks/JIVEtalk. Accessed:
01/06/2016.
Alina Leidinger (ICL) RJICA June 8, 2016 23 / 28
24/28
Toy Example: Robust JIVE Output
Alina Leidinger (ICL) RJICA June 8, 2016 24 / 28
25/28
Toy Example: JIVE Output for Occluded Images
Alina Leidinger (ICL) RJICA June 8, 2016 25 / 28
26/28
Toy Example: Robust JIVE Output for Occluded Images
Alina Leidinger (ICL) RJICA June 8, 2016 26 / 28
27/28
Toy Example: Robust JIVE Error Matrix
Alina Leidinger (ICL) RJICA June 8, 2016 27 / 28
28/28
Thank you!
Alina Leidinger (ICL) RJICA June 8, 2016 28 / 28

More Related Content

What's hot

L02 datacentres observables
L02 datacentres observablesL02 datacentres observables
L02 datacentres observablesajsatienza
 
IROS 2011 talk 2 (Filippo's file)
IROS 2011 talk 2 (Filippo's file)IROS 2011 talk 2 (Filippo's file)
IROS 2011 talk 2 (Filippo's file)Gianluca Antonelli
 
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)hirokazutanaka
 
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...hirokazutanaka
 

What's hot (7)

Tprimal agh
Tprimal aghTprimal agh
Tprimal agh
 
rinko2010
rinko2010rinko2010
rinko2010
 
L02 datacentres observables
L02 datacentres observablesL02 datacentres observables
L02 datacentres observables
 
IROS 2011 talk 2 (Filippo's file)
IROS 2011 talk 2 (Filippo's file)IROS 2011 talk 2 (Filippo's file)
IROS 2011 talk 2 (Filippo's file)
 
Underwater manipulation
Underwater manipulationUnderwater manipulation
Underwater manipulation
 
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
 
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...
 

Similar to Undergraduate Research Project "Robust Joint and Individual Component Analysis"

Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...
Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...
Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...Chiheb Ben Hammouda
 
Appendix of downlink coverage probability in heterogeneous cellular networks ...
Appendix of downlink coverage probability in heterogeneous cellular networks ...Appendix of downlink coverage probability in heterogeneous cellular networks ...
Appendix of downlink coverage probability in heterogeneous cellular networks ...Cora Li
 
7. toda yamamoto-granger causality
7. toda yamamoto-granger causality7. toda yamamoto-granger causality
7. toda yamamoto-granger causalityQuang Hoang
 
Trunsored data analysis
Trunsored data analysisTrunsored data analysis
Trunsored data analysisHideo Hirose
 
5. cem granger causality ecm
5. cem granger causality  ecm 5. cem granger causality  ecm
5. cem granger causality ecm Quang Hoang
 
Deep IRL by C language
Deep IRL by C languageDeep IRL by C language
Deep IRL by C languageMasato Nakai
 
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...Rene Kotze
 
離散値ベクトル再構成手法とその通信応用
離散値ベクトル再構成手法とその通信応用離散値ベクトル再構成手法とその通信応用
離散値ベクトル再構成手法とその通信応用Ryo Hayakawa
 
TSU Seminar, JNCASR, March 2016
TSU Seminar, JNCASR, March 2016TSU Seminar, JNCASR, March 2016
TSU Seminar, JNCASR, March 2016Amit Bhattacharjee
 
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...Chiheb Ben Hammouda
 
On the Exponential Diophantine Equation
On the Exponential Diophantine EquationOn the Exponential Diophantine Equation
On the Exponential Diophantine EquationIRJET Journal
 
Principal Component Analysis for Tensor Analysis and EEG classification
Principal Component Analysis for Tensor Analysis and EEG classificationPrincipal Component Analysis for Tensor Analysis and EEG classification
Principal Component Analysis for Tensor Analysis and EEG classificationTatsuya Yokota
 
Hidden Markov Models common probability formulas
Hidden Markov Models common probability formulasHidden Markov Models common probability formulas
Hidden Markov Models common probability formulasNidhal Selmi
 

Similar to Undergraduate Research Project "Robust Joint and Individual Component Analysis" (20)

Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...
Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...
Seminar Talk: Multilevel Hybrid Split Step Implicit Tau-Leap for Stochastic R...
 
Beamer 4.pdf
Beamer 4.pdfBeamer 4.pdf
Beamer 4.pdf
 
Appendix of downlink coverage probability in heterogeneous cellular networks ...
Appendix of downlink coverage probability in heterogeneous cellular networks ...Appendix of downlink coverage probability in heterogeneous cellular networks ...
Appendix of downlink coverage probability in heterogeneous cellular networks ...
 
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
 
CMU_13
CMU_13CMU_13
CMU_13
 
7. toda yamamoto-granger causality
7. toda yamamoto-granger causality7. toda yamamoto-granger causality
7. toda yamamoto-granger causality
 
Trunsored data analysis
Trunsored data analysisTrunsored data analysis
Trunsored data analysis
 
5. cem granger causality ecm
5. cem granger causality  ecm 5. cem granger causality  ecm
5. cem granger causality ecm
 
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
PMED Opening Workshop - Inference on Individualized Treatment Rules from Obse...
 
Deep IRL by C language
Deep IRL by C languageDeep IRL by C language
Deep IRL by C language
 
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...
NITheP UKZN Seminar: Prof. Alexander Gorokhov (Samara State University, Russi...
 
離散値ベクトル再構成手法とその通信応用
離散値ベクトル再構成手法とその通信応用離散値ベクトル再構成手法とその通信応用
離散値ベクトル再構成手法とその通信応用
 
TSU Seminar, JNCASR, March 2016
TSU Seminar, JNCASR, March 2016TSU Seminar, JNCASR, March 2016
TSU Seminar, JNCASR, March 2016
 
QMC Program: Trends and Advances in Monte Carlo Sampling Algorithms Workshop,...
QMC Program: Trends and Advances in Monte Carlo Sampling Algorithms Workshop,...QMC Program: Trends and Advances in Monte Carlo Sampling Algorithms Workshop,...
QMC Program: Trends and Advances in Monte Carlo Sampling Algorithms Workshop,...
 
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...
MCQMC 2020 talk: Importance Sampling for a Robust and Efficient Multilevel Mo...
 
On the Exponential Diophantine Equation
On the Exponential Diophantine EquationOn the Exponential Diophantine Equation
On the Exponential Diophantine Equation
 
From unconventional to extreme to functional materials.
From unconventional to extreme to functional materials.From unconventional to extreme to functional materials.
From unconventional to extreme to functional materials.
 
Principal Component Analysis for Tensor Analysis and EEG classification
Principal Component Analysis for Tensor Analysis and EEG classificationPrincipal Component Analysis for Tensor Analysis and EEG classification
Principal Component Analysis for Tensor Analysis and EEG classification
 
Hidden Markov Models common probability formulas
Hidden Markov Models common probability formulasHidden Markov Models common probability formulas
Hidden Markov Models common probability formulas
 
Geneticalgorithm
GeneticalgorithmGeneticalgorithm
Geneticalgorithm
 

Recently uploaded

(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Serviceranjana rawat
 
RadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfRadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfgstagge
 
How we prevented account sharing with MFA
How we prevented account sharing with MFAHow we prevented account sharing with MFA
How we prevented account sharing with MFAAndrei Kaleshka
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingNeil Barnes
 
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样vhwb25kk
 
RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998YohFuh
 
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...ThinkInnovation
 
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)jennyeacort
 
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDINTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDRafezzaman
 
9654467111 Call Girls In Munirka Hotel And Home Service
9654467111 Call Girls In Munirka Hotel And Home Service9654467111 Call Girls In Munirka Hotel And Home Service
9654467111 Call Girls In Munirka Hotel And Home ServiceSapana Sha
 
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfKantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfSocial Samosa
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptSonatrach
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfLars Albertsson
 
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptx
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptxAmazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptx
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptxAbdelrhman abooda
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPramod Kumar Srivastava
 
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...dajasot375
 
04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationshipsccctableauusergroup
 

Recently uploaded (20)

(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
 
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
 
RadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfRadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdf
 
How we prevented account sharing with MFA
How we prevented account sharing with MFAHow we prevented account sharing with MFA
How we prevented account sharing with MFA
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data Storytelling
 
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样
1:1定制(UQ毕业证)昆士兰大学毕业证成绩单修改留信学历认证原版一模一样
 
RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998
 
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...
Predictive Analysis - Using Insight-informed Data to Determine Factors Drivin...
 
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
 
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDINTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
 
9654467111 Call Girls In Munirka Hotel And Home Service
9654467111 Call Girls In Munirka Hotel And Home Service9654467111 Call Girls In Munirka Hotel And Home Service
9654467111 Call Girls In Munirka Hotel And Home Service
 
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfKantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
 
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdf
 
Call Girls in Saket 99530🔝 56974 Escort Service
Call Girls in Saket 99530🔝 56974 Escort ServiceCall Girls in Saket 99530🔝 56974 Escort Service
Call Girls in Saket 99530🔝 56974 Escort Service
 
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptx
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptxAmazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptx
Amazon TQM (2) Amazon TQM (2)Amazon TQM (2).pptx
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
 
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
 
04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships
 

Undergraduate Research Project "Robust Joint and Individual Component Analysis"

  • 1. 1/28 Robust Joint and Individual Component Analysis Alina Leidinger Imperial College London June 8, 2016 Alina Leidinger (ICL) RJICA June 8, 2016 1 / 28
  • 2. 2/28 Joint and Individual Variation Explained (JIVE) 1 2 3 4 1 http://www.uni-regensburg.de/Fakultaeten/phil_Fak_II/Psychologie/ Psy_II/beautycheck/english/durchschnittsgesichter/m(01-32)_gr.jpg. 2 http: //img13.deviantart.net/47f8/i/2014/177/5/a/sketch___male_face_by_pmucks- d7o53mf.png. 3 https://www.etc.cmu.edu/projects/atl/images/facial/plainface.jpg. 4 http://www.thermology.com/infrared22.JPG. Alina Leidinger (ICL) RJICA June 8, 2016 2 / 28
  • 3. 3/28 Contents Joint and Individual Variation Explained (JIVE) Robust JIVE Robust Nuclear Norm Regularised JIVE Toy Example Alina Leidinger (ICL) RJICA June 8, 2016 3 / 28
  • 4. 4/28 Joint and Individual Variation Explained (JIVE)5 different views on the same data e.g. measurements of different quantities on the same set of objects Original application: cancer research Data from The Cancer Genome Atlas Potential fields of Application: clinical studies (patient groups), Finance (markets), e-commerce (websites), environmental sciences (locations), facial recognition (expressions) ... 5 Eric F. Lock et al. “Joint and individual variation explained (JIVE) for integrated analysis of multiple data types”. In: Annals of Applied Statistics 7.1 (2013). ID: TN_scopus2-s2.0-84876058478, pp. 523–542. Alina Leidinger (ICL) RJICA June 8, 2016 4 / 28
  • 5. 5/28 JIVE suitability for high dimensional data n < pi Alina Leidinger (ICL) RJICA June 8, 2016 5 / 28
  • 6. 6/28 JIVE: Mathematical Framework Xi = Ji + Ai + Ei i = 1, . . . , k or    X1 ... Xk    X =    J1 ... Jk    J +    A1 ... Ak    A +    E1 ... Ek    E require uncorrelatedness between joint and individual components: JAT i = 0 Alina Leidinger (ICL) RJICA June 8, 2016 6 / 28
  • 7. 7/28 JIVE: Optimisation Problem arg min J,Ai X − J − [AT 1 , . . . , AT k ]T 2 F s.t. JAT i = 0, rank(J) = r, rank(Ai ) = ri J(t+1) = arg min J J − (X − [A (t)T 1 , . . . , A (t)T k ]T ) 2 F s.t. rank(J) = r A (t+1) i = arg min Ai Ai − (Xi − J (t+1) i ) 2 F s.t. rank(Ai ) = ri , J(t+1) AT i = 0 Alina Leidinger (ICL) RJICA June 8, 2016 7 / 28
  • 8. 8/28 JIVE: Alternating Optimisation J(t+1) = arg min J J − (X − [A (t)T 1 , . . . , A (t)T k ]T ) 2 F s.t. rank(J) = r → rank-r SVD of X − [A (t)T 1 , . . . , A (t)T k ]T A (t+1) i = arg min Ai Ai − (Xi − J (t+1) i ) 2 F s.t rank(Ai ) = ri , J(t+1) AT i = 0 → (rank-ri SVD of Xi − J (t+1) i ) × (I − VV T ) Singular Value Decomposition: J(t+1) = UΣV T Alina Leidinger (ICL) RJICA June 8, 2016 8 / 28
  • 9. 9/28 JIVE: Algorithm6 Algorithm 1 JIVE 1: Xjoint = X = [XT 1 . . . XT k ]T 2: while not converged do 3: Estimate J by a rank r SVD of Xjoint i.e. J = Ur Λr V T r 4: for i = 1, . . . , k do 5: Set XIndividual i = Xi − Ji 6: Estimate Ai by a rank ri SVD of XIndividual i (I − VV T ) 7: Set XJoint i = Xi − Ai 8: XJoint = [XJointT 1 . . . XJointT k ]T 6 Lock et al. “Supplement to "JOINT AND INDIVIDUAL VARIATION EXPLAINED (JIVE) FOR INTEGRATAED ANALYSIS OF MULTIPLE DATA TYPES"”. In: The Annals of Applied Statistics 7.1 (2013). Alina Leidinger (ICL) RJICA June 8, 2016 9 / 28
  • 10. 10/28 JIVE: Criticism sensitive to outliers, missing data points sensitive to gross non-Gaussian noise combinatorial pre-processing step: inefficient, unsuitable for analysis of images Alina Leidinger (ICL) RJICA June 8, 2016 10 / 28
  • 11. 11/28 Robust JIVE arg min J,Ai E 0 s.t. X = J + A + E, JAT i = 0 → arg min J,Ai E 1 s.t. X = J + A + E, JAT i = 0 non-Gaussian noise sparsity 0-norm → 1-norm7 7 Robert Tibshirani. “Regression Shrinkage and Selection via the Lasso”. In: Journal of the Royal Statistical Society.Series B (Methodological) 58.1 (1996). ID: TN_jstor_archive2346178, pp. 267–288. Alina Leidinger (ICL) RJICA June 8, 2016 11 / 28
  • 12. 12/28 Robust JIVE: Augmented Lagrangian Method8 arg max X f (X) s.t. g(X) = 0 X(t) = arg max X f (X) + Z(t) , g(X) + µ(t) 2 g(X) 2 F = arg max X f (X) + µ(t) 2 g(X) + Z(t) /µ(t) 2 F µ(t+1) = ρµ(t) Z(t+1) = Z(t) + µ(t) (g(X)) 8 Dimitri P. Bertsekas. Constrained optimization and Lagrange multiplier methods. Bibliography: p.383-392.- Includes index.; ID: 44IMP_ALMA_DS2145810210001591. New York ; London: Academic Press, 1982. Alina Leidinger (ICL) RJICA June 8, 2016 12 / 28
  • 13. 13/28 Robust JIVE: Alternating Direction Method9 arg min E E 1+ µ 2 X − J − A − E + Z/µ 2 F s.t. JAT i = 0, rank(J) = r, rank(Ai ) = ri E∗ = arg min E E 1+ µ 2 X − J − A − E + Z/µ 2 F J∗ = arg min J µ 2 X − J − A − E + Z/µ 2 F s.t. rank(J) = r A∗ i = arg min Ai µ 2 Xi − Ji − Ai − Ei + Zi /µ 2 F s.t rank(Ai ) = ri , JAT i = 0 9 Daniel Gabay and Bertrand Mercier. “A dual algorithm for the solution of nonlinear variational problems via finite element approximation”. In: Computers and Mathematics with Applications 2.1 (1976). ID: TN_sciversesciencedirect_elsevier0898-1221(76)90003-1, pp. 17–40. Alina Leidinger (ICL) RJICA June 8, 2016 13 / 28
  • 14. 14/28 Robust JIVE: ADM Steps10 E(t+1) = arg min E E 1+ µ(t) 2 X − J(t) − A(t) − E + Z(t) /µ(t) 2 F = Sµ(t)−1 (X − J(t) − A(t) + Z(t) /µ(t) ) Sa(x) =    x − a x > a x + a x < a 0 otherwise 10 Emmanuel Candès J. et al. “Robust principal component analysis?” In: Journal of the ACM (JACM) 58.3 (2011). ID: TN_acm1970395, pp. 1–37. Alina Leidinger (ICL) RJICA June 8, 2016 14 / 28
  • 15. 15/28 Robust JIVE: Algorithm Algorithm 2 Robust JIVE 1: Output: J, Ai , E 2: Xjoint = X = [XT 1 . . . XT k ]T 3: Initialise µ(0) = 0.01, Z(0) = 0, E(0) = 0, J(0) = 0, A (0) i = 0, ρ = 1.1 4: while not converged do 5: Set E(t+1) = Sµ(t)−1 (X − J(t) − A(t) + Z(t)/µ(t)) 6: Set J(t+1) equal to a rank r SVD of X − A(t) − E(t+1) + Z(t)/µ(t) 7: for i = 1, . . . , k do 8: Set A (t+1) i equal to a rank ri SVD of (Xi − J (t+1) i − E (t+1) i + Z (t) i /µ(t))(I − VV T ) where J(t+1) = UΣV T 9: Update Z(t+1) = Z(t) + µ(t)(X − J(t+1) − A(t+1) − E(t+1)) 10: Update µ(t+1) = µ(t)ρ 11: t = t + 1 Alina Leidinger (ICL) RJICA June 8, 2016 15 / 28
  • 16. 16/28 Robust Nuclear Norm Regularised JIVE minJ,Ai ,E αrank(J) + i αi rank(Ai ) + λ E 1 s.t. X = J + A + E, JAT i = 0 rank function → nuclear norm, . ∗ 11 11 Benjamin Recht, Maryam Fazel, and Pablo A. Parrilo. “Guaranteed Minimum- Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization”. In: SIAM Review 52.3 (2010). ID: TN_siam10.1137/070697835, pp. 471–501. Alina Leidinger (ICL) RJICA June 8, 2016 16 / 28
  • 17. 17/28 Robust Nuclear Norm Regularised JIVE min J,Ai ,E α J ∗+ i αi Ai ∗+λ E 1 s.t. X = J + A + E, JAT i = 0 or min J,Ai ,Ri ,E α J ∗+ i αi Ri ∗+λ E 1 s.t. X = J + A + E, Ai = Ri , JAT i = 0 = min J,Ai ,Ri ,E α J ∗+ i αi Ri ∗+λ E 1 + µ 2 X − J − A − E + Z µ 2 F + i γi 2 Ri − Ai + Yi γi 2 F s.t.JAT i = 0 Alina Leidinger (ICL) RJICA June 8, 2016 17 / 28
  • 18. 18/28 Robust Nuclear Norm Regularised JIVE: ADM Steps12 E(t+1) = arg min E λ E 1+ µ(t) 2 E − (X − J(t+1) − A(t+1) + Z(t) µ(t) ) 2 F = Sλ/µ(t) (X − J(t+1) − A(t+1) + Z(t) /µ(t) ) 12 Emmanuel Candès J. et al. “Robust principal component analysis?” In: Journal of the ACM (JACM) 58.3 (2011). ID: TN_acm1970395, pp. 1–37. Alina Leidinger (ICL) RJICA June 8, 2016 18 / 28
  • 19. 19/28 Robust Nuclear Norm Regularised JIVE13 J(t+1) = arg min J α J ∗+ µ(t) 2 J − (X − A(t) − E(t) + Z(t) µ(t) ) 2 F = Dα/µ(t) (X − A(t) − E(t) + Z(t) µ(t) ) R (t+1) i = arg min Ri αi Ri ∗+ γ (t) i 2 Ri − A (t) i + Y (t) i γ (t) i 2 F = Dαi /γ (t) i (A (t+1) i − Y (t) i γ (t) i ) Dλ(X) = Udiag(max{σi − λ, 0})V T (X = UΣV T ) 13 Jian-Feng Cai, Emmanuel J. Candès, and Zuowei Shen. “A Singular Value Thresholding Algorithm for Matrix Completion”. In: SIAM Journal on Optimization 20.4 (2010). ID: TN_siam10.1137/080738970, pp. 1956–1982. Alina Leidinger (ICL) RJICA June 8, 2016 19 / 28
  • 20. 20/28 Robust Nuclear Norm Regularised JIVE: Linearising Cost Function1415 arg min Ai µ 2 Xi − Ji − Ai − Ei + Zi µ 2 F + γi 2 Ri − Ai + Yi γi 2 F f (Ai ) s.t.JAT i = 0 f (Ai ) ≈ f (A (t) i ) + f (A (t) i ), Ai − A (t) i + 1 2τ Ai − A (t) i 2 F = f (A (t) i ) + 1 2τ Ai − A (t) i + τ f (A (t) i ) 2 F 14 Junfeng Yang and Xiaoming Yuan. “Linearized augmented Lagrangian and alternating direction methods for nuclear norm minimization”. In: Mathematics of Computation 82.281 (2013), pp. 301–329. 15 Zhouchen Lin, Risheng Liu, and Zhixun Su. “Linearized Alternating Direction Method with Adaptive Penalty for Low-Rank Representation”. In: (2011). ID: TN_arxiv1109.0367. Alina Leidinger (ICL) RJICA June 8, 2016 20 / 28
  • 21. 21/28 Robust Nuclear Norm Regularised JIVE: Linearising Cost Function A (t+1) i = arg min Ai 1 2τ Ai − A (t) i + τ f (A (t) i ) 2 F s.t. JAT i = 0 = (A (t) i − τ f (A (t) i ))(I − VV T ) J(t+1) = UΣV T f (A (t) i ) =µ(t) (A (t) i − Xi + J (t+1) i − Z (t) i /µ(t) ) + γ (t) i (A (t) i − R (t) i − Y (t) i /γ (t) i )) Alina Leidinger (ICL) RJICA June 8, 2016 21 / 28
  • 22. 22/28 Toy Example: Artworks 16 Alina Leidinger (ICL) RJICA June 8, 2016 22 / 28
  • 23. 23/28 Toy Example Figure: Superposed Images 17 17 Joint and Individual Variation Explained (JIVE) for the Integrated Analysis of Multiple Datatypes. http://www.tc.umn.edu/~elock/Talks/JIVEtalk. Accessed: 01/06/2016. Alina Leidinger (ICL) RJICA June 8, 2016 23 / 28
  • 24. 24/28 Toy Example: Robust JIVE Output Alina Leidinger (ICL) RJICA June 8, 2016 24 / 28
  • 25. 25/28 Toy Example: JIVE Output for Occluded Images Alina Leidinger (ICL) RJICA June 8, 2016 25 / 28
  • 26. 26/28 Toy Example: Robust JIVE Output for Occluded Images Alina Leidinger (ICL) RJICA June 8, 2016 26 / 28
  • 27. 27/28 Toy Example: Robust JIVE Error Matrix Alina Leidinger (ICL) RJICA June 8, 2016 27 / 28
  • 28. 28/28 Thank you! Alina Leidinger (ICL) RJICA June 8, 2016 28 / 28