beyond the graphical lasso structure learning via inverse
play

Beyond the graphical Lasso: Structure learning via inverse - PowerPoint PPT Presentation

Mar 12, 2024 •370 likes •1.36k views

Beyond the graphical Lasso: Structure learning via inverse covariance estimation Po-Ling Loh UC Berkeley Department of Statistics ICML Workshop on Covariance Selection and Graphical Model Structure Learning June 26, 2014 Joint work with

  1. Beyond the graphical Lasso: Structure learning via inverse covariance estimation Po-Ling Loh UC Berkeley Department of Statistics ICML Workshop on Covariance Selection and Graphical Model Structure Learning June 26, 2014 Joint work with Martin Wainwright (UC Berkeley) & Peter B¨ uhlmann (ETH Z¨ urich) P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 1 / 40

  2. Outline Introduction 1 Generalized inverse covariances 2 Linear structural equation models 3 Corrupted data 4 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 2 / 40

  3. Outline Introduction 1 Generalized inverse covariances 2 Linear structural equation models 3 Corrupted data 4 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 3 / 40

  4. Undirected graphical models Undirected graph G = ( V , E ) Joint distribution of ( X 1 , . . . , X p ), where | V | = p X 3 X 1 X 2 X p P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 4 / 40

  5. Undirected graphical models Undirected graph G = ( V , E ) Joint distribution of ( X 1 , . . . , X p ), where | V | = p X 3 X 1 X 2 X p Markov property: ( s , t ) / ∈ E = ⇒ X s ⊥ ⊥ X t | X \{ s , t } P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 4 / 40

  6. Undirected graphical models Undirected graph G = ( V , E ) Joint distribution of ( X 1 , . . . , X p ), where | V | = p A X 3 X 1 B X 2 X p S More generally, X A ⊥ ⊥ X B | X S when S ⊆ V separates A from B P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 4 / 40

  7. Directed graphical models Directed acyclic graph G = ( V , E ) X 1 X 2 X 3 X p Markov property: X j ⊥ ⊥ X Nondesc( j ) | X Pa( j ) , ∀ j P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 5 / 40

  8. Structure learning � � n ( X ( i ) 1 , X ( i ) 2 , . . . , X ( i ) Goal: Edge recovery from n samples: p ) i =1 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 6 / 40

  9. Structure learning � � n ( X ( i ) 1 , X ( i ) 2 , . . . , X ( i ) Goal: Edge recovery from n samples: p ) i =1 High-dimensional setting: p ≫ n , assume deg( G ) ≤ d P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 6 / 40

  10. Structure learning � � n ( X ( i ) 1 , X ( i ) 2 , . . . , X ( i ) Goal: Edge recovery from n samples: p ) i =1 High-dimensional setting: p ≫ n , assume deg( G ) ≤ d Sources of corruption: non-i.i.d. observations, contamination by noise/missing data P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 6 / 40

  11. Structure learning � � n ( X ( i ) 1 , X ( i ) 2 , . . . , X ( i ) Goal: Edge recovery from n samples: p ) i =1 High-dimensional setting: p ≫ n , assume deg( G ) ≤ d Sources of corruption: non-i.i.d. observations, contamination by noise/missing data Note: Structure learning generally harder for directed graphs (topological order unknown) P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 6 / 40

  12. Graphical Lasso When ( X 1 , . . . , X p ) ∼ N (0 , Σ), well-known fact: (Σ − 1 ) st = 0 ⇐ ⇒ ( s , t ) / ∈ E P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 7 / 40

  13. Graphical Lasso When ( X 1 , . . . , X p ) ∼ N (0 , Σ), well-known fact: (Σ − 1 ) st = 0 ⇐ ⇒ ( s , t ) / ∈ E Establishes statistical consistency of graphical Lasso (Yuan & Lin ’07):     � �  trace( � Θ ∈ arg min ΣΘ) − log det(Θ) + λ | Θ st |  Θ � 0 s � = t P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 7 / 40

  14. Some observations P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 8 / 40

  15. Some observations Only sample-based quantity is � Σ:     � �  trace( � Θ ∈ arg min ΣΘ) − log det(Θ) + λ | Θ st |  Θ � 0 s � = t P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 8 / 40

  16. Some observations Only sample-based quantity is � Σ:     � �  trace( � Θ ∈ arg min ΣΘ) − log det(Θ) + λ | Θ st |  Θ � 0 s � = t Although graphical Lasso is penalized Gaussian MLE , can always be used to estimate � Θ from � Σ: � � (Σ ∗ ) − 1 = arg min trace(Σ ∗ Θ) − log det(Θ) Θ P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 8 / 40

  17. Some observations Only sample-based quantity is � Σ:     � �  trace( � Θ ∈ arg min ΣΘ) − log det(Θ) + λ | Θ st |  Θ � 0 s � = t Although graphical Lasso is penalized Gaussian MLE , can always be used to estimate � Θ from � Σ: � � (Σ ∗ ) − 1 = arg min trace(Σ ∗ Θ) − log det(Θ) Θ We extend graphical Lasso to discrete-valued data (undirected case) and linear structural equation models (directed case) P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 8 / 40

  18. Theory for graphical Lasso If � � log p log p � � Σ − Σ ∗ � max � and λ � , n n then �� � log p � � Θ − Θ ∗ � max � + λ n P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 9 / 40

  19. Theory for graphical Lasso If � � log p log p � � Σ − Σ ∗ � max � and λ � , n n then �� � log p � � Θ − Θ ∗ � max � + λ n Deviation condition holds w.h.p. for various ensembles (e.g., sub-Gaussian) P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 9 / 40

  20. Theory for graphical Lasso If � � log p log p � � Σ − Σ ∗ � max � and λ � , n n then �� � log p � � Θ − Θ ∗ � max � + λ n Deviation condition holds w.h.p. for various ensembles (e.g., sub-Gaussian) � Thresholding � log p Θ at level yields correct support n P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 9 / 40

  21. Outline Introduction 1 Generalized inverse covariances 2 Linear structural equation models 3 Corrupted data 4 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 10 / 40

  22. Non-Gaussian distributions (Liu et al. ’09, ’12): ( X 1 , . . . , X p ) follows nonparanormal distribution if ( f 1 ( X 1 ) , . . . , f p ( X p )) ∼ N (0 , Σ), and f j ’s monotone and differentiable P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 11 / 40

  23. Non-Gaussian distributions (Liu et al. ’09, ’12): ( X 1 , . . . , X p ) follows nonparanormal distribution if ( f 1 ( X 1 ) , . . . , f p ( X p )) ∼ N (0 , Σ), and f j ’s monotone and differentiable Then ( i , j ) / ∈ E iff Θ ij = 0 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 11 / 40

  24. Non-Gaussian distributions (Liu et al. ’09, ’12): ( X 1 , . . . , X p ) follows nonparanormal distribution if ( f 1 ( X 1 ) , . . . , f p ( X p )) ∼ N (0 , Σ), and f j ’s monotone and differentiable Then ( i , j ) / ∈ E iff Θ ij = 0 In general non-Gaussian setting, relationship between entries of Θ = Σ − 1 and edges of G unknown P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 11 / 40

  25. Discrete graphical models Assume X i ’s take values in a discrete set: { 0 , 1 , . . . , m − 1 } P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 12 / 40

  26. Discrete graphical models Assume X i ’s take values in a discrete set: { 0 , 1 , . . . , m − 1 } Our results: Establish relationship between augmented inverse covariance matrices and edge structure New algorithms for structure learning in discrete graphs P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 12 / 40

  27. An illustrative example Binary Ising model:   � �  , P θ ( x 1 , . . . , x p ) ∝ exp θ s x s + θ st x s x t s ∈ V ( s , t ) ∈ E P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 13 / 40

  28. An illustrative example Binary Ising model:   � �  , P θ ( x 1 , . . . , x p ) ∝ exp θ s x s + θ st x s x t s ∈ V ( s , t ) ∈ E θ ∈ R p + ( p 2 ) , ( x 1 , . . . , x p ) ∈ { 0 , 1 } p P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 13 / 40

  29. An illustrative example Ising models with θ s = 0 . 1 , θ st = 2 X 1 X 4 X 2 X 3 X 1 X 4 X 2 X 3 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 14 / 40

  30. An illustrative example Ising models with θ s = 0 . 1 , θ st = 2   X 1 X 4 9 . 80 − 3 . 59 0 0   − 3 . 59 34 . 30 − 4 . 77 0   Θ chain =   0 − 4 . 77 34 . 30 − 3 . 59 0 0 − 3 . 59 9 . 80 X 2 X 3   X 1 X 4 51 . 37 − 5 . 37 − 0 . 17 − 5 . 37   − 5 . 37 51 . 37 − 5 . 37 − 0 . 17   Θ loop =   − 0 . 17 − 5 . 37 51 . 37 − 5 . 37 − 5 . 37 − 0 . 17 − 5 . 37 51 . 37 X 2 X 3 P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 14 / 40

  31. An illustrative example Ising models with θ s = 0 . 1 , θ st = 2   X 1 X 4 9 . 80 − 3 . 59 0 0   − 3 . 59 34 . 30 − 4 . 77 0   Θ chain =   0 − 4 . 77 34 . 30 − 3 . 59 0 0 − 3 . 59 9 . 80 X 2 X 3   X 1 X 4 51 . 37 − 5 . 37 − 0 . 17 − 5 . 37   − 5 . 37 51 . 37 − 5 . 37 − 0 . 17   Θ loop =   − 0 . 17 − 5 . 37 51 . 37 − 5 . 37 − 5 . 37 − 0 . 17 − 5 . 37 51 . 37 X 2 X 3 Θ is graph-structured for chain, but not loop P. Loh (UC Berkeley) Beyond the graphical Lasso June 26, 2014 14 / 40

Recommend

Probabilistic Graphical Models Probabilistic Graphical Models Structure learning in Bayesian

Probabilistic Graphical Models Probabilistic Graphical Models Structure learning in Bayesian

Probabilistic Graphical Models Probabilistic Graphical Models Structure learning in Bayesian networks Siamak Ravanbakhsh Fall 2019 Learning objectives Learning objectives why structure learning is hard? two approaches to structure learning

1.04k views • 74 slides
Bilevel Learning of the Group Lasso Structure Jordan Frecon 1 , Saverio Salzo 1 , Massimiliano

Bilevel Learning of the Group Lasso Structure Jordan Frecon 1 , Saverio Salzo 1 , Massimiliano

Bilevel Learning of the Group Lasso Structure Jordan Frecon 1 , Saverio Salzo 1 , Massimiliano Pontil 1 , 2 1 CSML - Istituto Italiano di Tecnologia 2 Dept of Computer Science - University College London Thirty-second Conference on Neural

327 views • 9 slides
Structure' Learning' Daphne Koller Why Structure Learning To learn model for new queries,

Structure' Learning' Daphne Koller Why Structure Learning To learn model for new queries,

Learning' Probabilis2c' Graphical' BN'Structure' Models' Structure' Learning' Daphne Koller Why Structure Learning To learn model for new queries, when domain expertise is not perfect For structure discovery, when inferring network

752 views • 61 slides
Learning in Graphical Models Andrea Passerini passerini@disi.unitn.it Machine Learning Learning

Learning in Graphical Models Andrea Passerini passerini@disi.unitn.it Machine Learning Learning

Learning in Graphical Models Andrea Passerini passerini@disi.unitn.it Machine Learning Learning in Graphical Models Learning graphical models Parameter estimation We assume the structure of the model is given We are given a dataset of

268 views • 23 slides
Learning Bayesian network : Given structure and completely observed data Probabilistic Graphical

Learning Bayesian network : Given structure and completely observed data Probabilistic Graphical

Learning Bayesian network : Given structure and completely observed data Probabilistic Graphical Models Sharif University of Technology Spring 2017 Soleymani Learning problem Target: true distribution that maybe correspond to

766 views • 61 slides
Structure Learning: the good, the bad, the ugly Graphical Model 10708 Carlos Guestrin

Structure Learning: the good, the bad, the ugly Graphical Model 10708 Carlos Guestrin

Koller & Friedman Chapter 13 Structure Learning: the good, the bad, the ugly Graphical Model 10708 Carlos Guestrin Carnegie Mellon University October 24 th , 2005 Project feedback by e-mail soon Announcements Where are we?

760 views • 38 slides
1. Algorithms for Inverse Reinforcement Learning 2. Apprenticeship learning via Inverse

1. Algorithms for Inverse Reinforcement Learning 2. Apprenticeship learning via Inverse

1. Algorithms for Inverse Reinforcement Learning 2. Apprenticeship learning via Inverse Reinforcement Learning Algorithms for Inverse Reinforcement Learning Andrew Ng and Stuart Russell Motivation Given: (1) measurements of an agent's

379 views • 34 slides
Learning in Graphical Models Problem Dimensions Model Bayes Nets Markov Nets

Learning in Graphical Models Problem Dimensions Model Bayes Nets Markov Nets

Learning in Graphical Models Problem Dimensions Model Bayes Nets Markov Nets Structure Known Unknown (structure learning) Data Complete Incomplete (missing values or hidden variables) Expectation-Maximization

814 views • 28 slides
Learning the Structure of Mixed Graphical Models Jason Lee with Trevor Hastie, Michael Saunders,

Learning the Structure of Mixed Graphical Models Jason Lee with Trevor Hastie, Michael Saunders,

Learning the Structure of Mixed Graphical Models Jason Lee with Trevor Hastie, Michael Saunders, Yuekai Sun, and Jonathan Taylor Institute of Computational & Mathematical Engineering Stanford University June 26th, 2014 Examples of

581 views • 45 slides
ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale

ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale

ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale Learning Topics Bayes Nets (Finish) Parameter Learning Structure Learning Readings: KF 18.1, 18.3; Barber 9.5, 10.4 Dhruv Batra Virginia

351 views • 32 slides
Probabilistic Graphical Models Probabilistic Graphical Models introduction to learning Siamak

Probabilistic Graphical Models Probabilistic Graphical Models introduction to learning Siamak

Probabilistic Graphical Models Probabilistic Graphical Models introduction to learning Siamak Ravanbakhsh Fall 2019 Learning objectives Learning objectives different goals of learning a graphical model effect of goals on the learning setup

273 views • 23 slides
Autonomous Navigation CSE 571 Inverse Optimal Control (Inverse Reinforcement Learning) Many

Autonomous Navigation CSE 571 Inverse Optimal Control (Inverse Reinforcement Learning) Many

5/25/20 Autonomous Navigation CSE 571 Inverse Optimal Control (Inverse Reinforcement Learning) Many slides by Drew Bagnell Carnegie Mellon University 1 2 Optimal Control Solution X X Learning Y Y Cost Map (Sensor Data) (Input)

489 views • 10 slides
Screening Rules for Lasso with Non-Convex Sparse Regularizers Joseph Salmon

Screening Rules for Lasso with Non-Convex Sparse Regularizers Joseph Salmon

Screening Rules for Lasso with Non-Convex Sparse Regularizers Joseph Salmon http://josephsalmon.eu Universit de Montpellier Joint work with A. Rakotomamonjy and G. Gasso 1 / 18 Motivation and objective Lasso and screening Learning

440 views • 24 slides
Graphical Models Independence & Factorization Including structure Complexity of

Graphical Models Independence & Factorization Including structure Complexity of

Graphical Models slides adopted from Sandro Schnborn 1 Graphical Models Independence & Factorization Including structure Complexity of multivariate problems Independence assumptions Graphical Models Graphs to depict

385 views • 23 slides
ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale

ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale

ECE 6504: Advanced Topics in Machine Learning Probabilistic Graphical Models and Large-Scale Learning Topics Bayes Nets (Finish) Structure Learning Readings: KF 18.4; Barber 9.5, 10.4 Dhruv Batra Virginia Tech Administrativia HW1

596 views • 30 slides
Probabilistic Graphical Models Probabilistic Graphical Models Parameter learning in Bayesian

Probabilistic Graphical Models Probabilistic Graphical Models Parameter learning in Bayesian

Probabilistic Graphical Models Probabilistic Graphical Models Parameter learning in Bayesian networks Siamak Ravanbakhsh Fall 2019 Learning objectives Learning objectives likelihood function and MLE role of the sufficient statistics MLE for

644 views • 60 slides
Ehsan Nazerfard nazerfard@eecs.wsu.edu October 11, 2011 Introduction: Graphical Models

Ehsan Nazerfard nazerfard@eecs.wsu.edu October 11, 2011 Introduction: Graphical Models

Ehsan Nazerfard nazerfard@eecs.wsu.edu October 11, 2011 Introduction: Graphical Models Tutorial: Bayesian Networks Structure Learning Approaches RAI: Recursive Autonomy Identification for Bayesian Network Structure Learning Bayesian

306 views • 29 slides
Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from

Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from

Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from Observations Daniel Brown*, Wonjoon Goo*, Prabhat Nagarajan, and Scott Niekum Inverse Reinforcement Learning Current approaches 1. Cant do better

289 views • 27 slides
Comparing Bayesian Networks and Structure Learning Algorithms (and other graphical models) Marco

Comparing Bayesian Networks and Structure Learning Algorithms (and other graphical models) Marco

Comparing Bayesian Networks and Structure Learning Algorithms (and other graphical models) Marco Scutari marco.scutari@stat.unipd.it Department of Statistical Sciences University of Padova October 20, 2009 Marco Scutari University of Padova

814 views • 34 slides
Lecture 1: Introduction Statistical and Computational Methods for Learning through Graphical

Lecture 1: Introduction Statistical and Computational Methods for Learning through Graphical

Lecture 1: Introduction Statistical and Computational Methods for Learning through Graphical Models (aka Probabilistic Graphical Models) BIOSTAT 830 September 6 th , 2016 Zhenke Wu Some materials adapted from Eric Xings CMU Graphical Model

478 views • 18 slides
Probabilistic Graphical Models Probabilistic Graphical Models Learning with partial observations

Probabilistic Graphical Models Probabilistic Graphical Models Learning with partial observations

Probabilistic Graphical Models Probabilistic Graphical Models Learning with partial observations Siamak Ravanbakhsh Fall 2019 Learning objectives Learning objectives different types of missing data learning with missing data and hidden vars:

1.04k views • 80 slides
Harnessing Structure in Optimization for Machine Learning Franck Iutzeler LJK, Univ. Grenoble

Harnessing Structure in Optimization for Machine Learning Franck Iutzeler LJK, Univ. Grenoble

Harnessing Structure in Optimization for Machine Learning Franck Iutzeler LJK, Univ. Grenoble Alpes Optimization for Machine Learning CIRM 9-13 March 2020 >>> Regularization in Learning Structure Regularization Linear inverse

507 views • 33 slides
A practical tour of optimization algorithms for the Lasso Alexandre Gramfort

A practical tour of optimization algorithms for the Lasso Alexandre Gramfort

A practical tour of optimization algorithms for the Lasso Alexandre Gramfort alexandre.gramfort@inria.fr Inria, Parietal Team Universit Paris-Saclay Huawei - Apr. 2017 Outline What is the Lasso Lasso with an orthogonal design

560 views • 29 slides
Inverse Folding and Sequence-Structure Maps of Ribonucleic Acids Peter Schuster Institut fr

Inverse Folding and Sequence-Structure Maps of Ribonucleic Acids Peter Schuster Institut fr

Inverse Folding and Sequence-Structure Maps of Ribonucleic Acids Peter Schuster Institut fr Theoretische Chemie und Molekulare Strukturbiologie der Universitt Wien Inverse Problem Workshop IPAM, UCLA, 22.10.2003 Web-Page for further

1.18k views • 97 slides

More recommend