Bernard Ng is a postdoctoral fellow under the Department of Statistics at the University of British Columbia. He is also affiliated with the Centre for Molecular Medicine and Therapeutics. Prior to his current position, he was a joint postdoctoral fellow at Stanford University and INRIA. He completed his MASc and PhD in Electrical Engineering at the University of British Columbia and his BASc in Electronics Engineering at Simon Fraser University. Recently, he expanded into the area of imaging genomics in investigating the interactions between genomic layers in relation to brain and behaviour
Bernard Ng
Sanmi (Oluwasanmi) Koyejo
Sanmi Koyejo an engineering research associate in the Poldrack Lab at Stanford University and an Assistant Professor (starting Fall 2016) in the Department of Computer Science at the University of Illinois at Urbana-Champaign. Koyejo’s research involves the development and analysis of principled methods for elucidating patterns in neuroimaging, genetics and other large-scale biological data for clinical applications and to advance scientific exploration. His current interests include the development of new tools for mapping time varying human brain networks and their association to behavioral and genetic factors.
Sandro Vega Pons
Sandro Vega-Pons is a post-doctoral fellow at NeuroInformatics Laboratory (NILab), a joint research unit between Fondazione Bruno Kessler (FBK), and University of Trento. He is also affiliated with the Pattern Analysis and Computer Vision (PAVIS) group of the Italian Institute of Technologies. He studied Computer Science at University of Havana, Cuba and received the PhD degree in Applied Mathematics from the same institution in 2011. His research interests include machine learning, pattern recognition, data mining, high dimensional statistics, complex networks and neuroimaging data analysis, covering from theoretical to practical aspects.
PRNI 2016 - TUTORIAL 1
Brain Network Analysis
Brain graph analysis provides a new suite of tools for the scientific study of brain connectivity and its association to behaviour, physiology, and diseases. Accumulating evidence is suggesting that neuropsychiatric disorders are linked to abnormal brain connectivity, and inter-subject differences in behaviour can be partially explained by variability in brain connectivity. This tutorial will cover a range of classic and modern methods for estimation, classification, and inference on brain graphs. The six main areas of focuses are: (i) estimation of connectivity graphs, (ii) classification using brain connectivity, (iii) statistical inference to identify relevant brain connections, (iv) kernels methods for brain graph inference, and (v) estimation and evaluation of time varying brain connectivity. Each section will include hands-on examples.
MONDAY, JUNE 20th
08:30 :: Registration
09:00 :: Introduction to Brain Graph Estimation, Classification, and Inference
- Motivation for brain graph analysis
- Overview of fMRI data and confounds
- Definition of estimation, classification, and inference
- Brief discussion of brain node definition
11:00 :: Brain Graph Inference
10:30 :: Coffee Break
12:30 :: Lunch Break
15:00 :: Coffee Break
15:30 :: Inference and estimation for switching networks
- Limitations of Pearson’s correlation in small sample regime,
i.e. ill-conditioned, non-positive definite, and mixing of direct and indirect interactions
- Partial correlation to distinguish direct and indirect effects
- L2 shrinkage estimators (e.g. Oracle Approximating Shrinkage) to deal with above limitations
- L1 shrinkage estimators (e.g. graphical LASSO) to deal with above limitations
- Bimodal integration of fMRI and diffusion MRI data for connectivity estimation
- Validation strategies
09:15 :: Brain Graph Estimation
10:00 :: Brain Graph Classification
- Sources of correlations between connectivity features, the positive semidefinite constraint on covariance matrices
- Problems with correlated features
- Tools from Riemannian geometry for removing the positive semidefinite constraint
- Complications with direct application of operators on Riemannian manifolds and its resolution
- Statistical inference for Pearson’s correlation and partial correlation at single subject level
- Statistical inference for Pearson’s correlation and partial correlation at group level
- Statistical inference for Pearson’s correlation and partial correlation on group contrast
- Statistical inference on classifier weights derived from e.g. LASSO model and SVM
- Techniques covered include: Bonferroni correction, independent and dependent FDR correction, stability selection,
p-value approximation for LASSO, SLOPE, Knockoff filter, max-t permutation test, and bootstrapped permutation test
11:30 :: Review of methods for two -sample inference with brain graph
- Notation and motivations
- Node correspondence property
- Classification based approach
- Hypothesis testing based approach
11:45 :: Kernel-based approach for brain graph discrimination
- Graph Kernels
- Graph embeddings
- Their applicability according to the one-to-one node correspondence property
13:30 :: Classification-based test versus Kernel two-sample test
- Classification accuracy as test statistic - Classification based test
- Maximum mean discrepancy as test statistic – Kernel two-sample test
- Comparison under different circumstances
- Comparison to non-kernel methods
14:00 :: Inference and estimation for slow varying networks
- Sliding windows and smoothed moving averages
- Estimators: Pearson, MTD, Partial Correlation
- New results on inference and estimation
- Known failure cases & open problems
- Discrete Hidden Markov models
- Comparison to two step k-means
- Continuous Hidden Markov models
- Known failure cases & open problems
Prerequisites
-
Some exposure to fMRI data analysis
-
Some experience with programming in MATLAB and Python
-
Some background in basic statistics, e.g. t-test, and machine learning, e.g. SVM
For additional information you may contact tutorial1@prni.org