Limits on Testing Structural Changes in Ising Models

We present novel information-theoretic limits on detecting sparse changes in Ising models, a problem that arises in many applications where network changes can occur due to some external stimuli. We show that the sample complexity for detecting sparse changes, in a minimax sense, is no better than learning the entire model even in settings with local sparsity. This is a surprising fact in light of prior work rooted in sparse recovery methods, which suggest that sample complexity in this context scales only with the number of network changes. To shed light on when change detection is easier than structured learning, we consider testing of edge deletion in forest-structured graphs, and high-temperature ferromagnets as case studies. We show for these that testing of small changes is similarly hard, but testing of \emph{large} changes is well-separated from structure learning. These results imply that testing of graphical models may not be amenable to concepts such as restricted strong convexity leveraged for sparsity pattern recovery, and algorithm development instead should be directed towards detection of large changes.

[1]  Guy Bresler,et al.  Learning a Tree-Structured Ising Model in Order to Make Predictions , 2016, The Annals of Statistics.

[2]  Ali Shojaie,et al.  Differential network analysis: A statistical perspective , 2020, Wiley interdisciplinary reviews. Computational statistics.

[3]  Constantinos Daskalakis,et al.  Testing Ising Models , 2016, IEEE Transactions on Information Theory.

[4]  T. Cai,et al.  Differential Markov random field analysis with an application to detecting differential microbial community networks. , 2019, Biometrika.

[5]  BY YIN XIA,et al.  Testing differential networks with applications to the detection of gene-gene interactions , 2015 .

[6]  Han Liu,et al.  High-temperature structure detection in ferromagnets , 2018, Information and Inference: A Journal of the IMA.

[7]  Daniel M. Kane,et al.  Testing Bayesian Networks , 2016, IEEE Transactions on Information Theory.

[8]  Martin J. Wainwright,et al.  A unified framework for high-dimensional analysis of $M$-estimators with decomposable regularizers , 2009, NIPS.

[9]  Alexandros G. Dimakis,et al.  On the Information Theoretic Limits of Learning Ising Models , 2014, NIPS.

[10]  Hong Yan,et al.  DiffNetFDR: differential network analysis with false discovery rate control , 2019, Bioinform..

[11]  R. Adamczak,et al.  A note on concentration for polynomials in the Ising model , 2018, Electronic Journal of Probability.

[12]  Guy Bresler,et al.  Efficiently Learning Ising Models on Arbitrary Graphs , 2014, STOC.

[13]  S. Fields,et al.  Protein-protein interactions: methods for detection and analysis , 1995, Microbiological reviews.

[14]  Y. Peres,et al.  Concentration inequalities for polynomials of contracting Ising models , 2017, 1706.00121.

[15]  Alexandros G. Dimakis,et al.  Sparse Logistic Regression Learns All Discrete Pairwise Graphical Models , 2018, NeurIPS.

[16]  T. Ideker,et al.  Differential network biology , 2012, Molecular systems biology.

[17]  Afonso S. Bandeira,et al.  Random Laplacian Matrices and Convex Relaxations , 2015, Found. Comput. Math..

[18]  Guy Bresler,et al.  Stein’s method for stationary distributions of Markov chains and application to Ising models , 2017, The Annals of Applied Probability.

[19]  Eric Vigoda,et al.  Lower bounds for testing graphical models: colorings and antiferromagnetic Ising models , 2019, COLT.

[20]  Weijian Yang,et al.  In vivo imaging of neural activity , 2017, Nature Methods.

[21]  Robert B. Griffiths,et al.  Rigorous Results for Ising Ferromagnets of Arbitrary Spin , 1969 .

[22]  Constantinos Daskalakis,et al.  Concentration of Multilinear Functions of the Ising Model with Applications to Network Data , 2017, NIPS.

[23]  Masashi Sugiyama,et al.  Support Consistency of Direct Sparse-Change Learning in Markov Networks , 2015, AAAI.

[24]  Volkan Cevher,et al.  On the Difficulty of Selecting Ising Models With Approximate Recovery , 2016, IEEE Transactions on Signal and Information Processing over Networks.

[25]  Constantinos Daskalakis,et al.  Learning and Testing Causal Models with Interventions , 2018, NeurIPS.

[26]  Venkatesh Saligrama,et al.  Lower bounds for two-sample structural change detection in ising and Gaussian models , 2017, 2017 55th Annual Allerton Conference on Communication, Control, and Computing (Allerton).

[27]  Aditya Guntuboyina Lower Bounds for the Minimax Risk Using $f$-Divergences, and Applications , 2011, IEEE Transactions on Information Theory.

[28]  Arindam Banerjee,et al.  Generalized Direct Change Estimation in Ising Model Structure , 2016, ICML.

[29]  Gary D Bader,et al.  The Genetic Landscape of a Cell , 2010, Science.

[30]  Masashi Sugiyama,et al.  Direct Learning of Sparse Changes in Markov Networks by Density Ratio Estimation , 2013, Neural Computation.

[31]  Grace L. Yang,et al.  Festschrift for Lucien Le Cam: Research Papers in Probability and Statistics. , 1997 .

[32]  Martin J. Wainwright,et al.  Information-Theoretic Limits of Selecting Binary Graphical Models in High Dimensions , 2009, IEEE Transactions on Information Theory.

[33]  Michael Chertkov,et al.  Optimal structure and parameter learning of Ising models , 2016, Science Advances.

[34]  Constantinos Daskalakis,et al.  Which Distribution Distances are Sublinearly Testable? , 2017, Electron. Colloquium Comput. Complex..

[35]  C. Pouet Nonparametric Goodness-of-Fit Testing Under Gaussian Models , 2004 .

[36]  Mark E. Bucklin,et al.  An integrative approach for analyzing hundreds of neurons in task performing mice using wide-field calcium imaging , 2016, Scientific Reports.

[37]  Guy Bresler,et al.  Optimal Single Sample Tests for Structured versus Unstructured Network Data , 2018, COLT.

[38]  Han Liu,et al.  Property testing in high-dimensional Ising models , 2017, The Annals of Statistics.

[39]  Mladen Kolar,et al.  Two‐sample inference for high‐dimensional Markov networks , 2019, Journal of the Royal Statistical Society: Series B (Statistical Methodology).

[40]  Isabelle Guyon,et al.  First Connectomics Challenge: From Imaging to Connectivity , 2015, Neural Connectomics.

[41]  Marloes H. Maathuis,et al.  Structure Learning in Graphical Modeling , 2016, 1606.02359.

[42]  Matthew B. Blaschko,et al.  Testing for Differences in Gaussian Graphical Models: Applications to Brain Connectivity , 2015, NIPS.

[43]  T. Cai,et al.  Direct estimation of differential networks. , 2014, Biometrika.

[44]  Venkatesh Saligrama,et al.  Two-Sample Testing can be as Hard as Structure Learning in Ising Models: Minimax Lower Bounds , 2018, 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[45]  Bin Yu Assouad, Fano, and Le Cam , 1997 .

[46]  A. Nobel,et al.  A TESTING BASED APPROACH TO THE DISCOVERY OF DIFFERENTIALLY CORRELATED VARIABLE SETS. , 2015, The annals of applied statistics.

[47]  Kenji Fukumizu,et al.  Learning sparse structural changes in high-dimensional Markov networks , 2017, Behaviormetrika.