Testing for Differences in Gaussian Graphical Models: Applications to Brain Connectivity
Authors: Eugene Belilovsky, Gaël Varoquaux, Matthew B. Blaschko
NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the techniques proposed on a set of synthetic examples as well as neuro-imaging dataset created for the study of autism. and 4 Experiments 4.1 Simulations We generate synthetic data based on two Gaussian graphical models with 75 vertices. ... We report the false positive rate, the power, the coverage and interval length as per [30] for the difference of graphs. ... 4.2 Autism Dataset Correlations in brain activity measured via f MRI reveal functional interactions between remote brain regions [18]. |
| Researcher Affiliation | Academia | Eugene Belilovsky 1,2,3, Gael Varoquaux2, Matthew Blaschko3 1University of Paris-Saclay, 2INRIA, 3KU Leuven {eugene.belilovsky, gael.varoquaux } @inria.fr matthew.blaschko@esat.kuleuven.be |
| Pseudocode | Yes | Algorithm 1 Difference Network Selection with Neighborhood Debiased Lasso and Algorithm 2 Difference Network Selection with Neighborhood Debiased Fused Lasso |
| Open Source Code | No | The paper does not provide an explicit statement or link to their own open-source code for the methodology described. It mentions using third-party packages like Mosek QP solver and hdi. |
| Open Datasets | Yes | The ABIDE (Autism Brain Imaging Data Exchange) dataset [8] gathers rest-f MRI from 1 112 subjects across, with 539 individuals suffering from autism spectrum disorder and 573 typical controls. We use the preprocessed and curated data1. 1http://preprocessed-connectomes-project.github.io/abide/ |
| Dataset Splits | Yes | We select c by 3fold cross validation k = {0.1, ..100} and The significance level is 5%, n1 = 800 and n2 = 60. All methods have false positive below the significance level and the debiased fused lasso dominates in terms of power. The coverage of the difference support and non-difference support is also best for the debiased fused lasso, which simultaneously has smaller confidence intervals on average. and Here we permute the two conditions (e.g. autism and control group) to compute a p-value and compare it to our test statistics. This provides us with a finite sample strict control on the error rate: a non-parametric validation of our parametric test. |
| Hardware Specification | No | The paper does not explicitly state the specific hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions 'Mosek QP solver package' and 'R package hdi' but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | We use the debiased lasso of [16] where we set λ = kˆσ p log p/n. We select c by 3fold cross validation k = {0.1, ..100} and M as prescribed in [16] which we obtain by solving a quadratic program. ˆσ is an unbiased estimator of the noise variance. For the debiased lasso we let both λ1 = k1 ˆσ2 p log p/n2 and λ2 = k2 ˆσ2 p log p/n2, and select based on 3-fold cross-validation from the same range as k. M1 and M2 are obtained as in Equation (13) with the bounds (14) being set with c = a = 2, sd = 2, s1,2 = 15, m = 0.01, and the cross validated λ1 and λ2. |