en
0.25
0.5
0.75
1.25
1.5
1.75
2
Resampling Based Methods for Design and Evaluation of Neurotechnology
Published on Dec 03, 20122340 Views
Brain imaging by PET, MR, EEG, and MEG has become a cornerstone in systems level neuroscience. Statistical analyses of neuroimage datasets face many interesting challenges including non-linearity and
Related categories
Chapter list
Resampling based methods for design and evaluation of neuroimaging models00:00
OUTLINE01:16
Recent reviews03:44
Functional MRI04:46
Main message: Models should be predictive and informative07:28
BOLD hemodynamics R-Bayes model selection12:16
Machine learning for neuroimage data14:02
Multivariate neuroimaging models18:40
AIM I: Generalizability22:41
Bias-variance trade-off as function of PCA dimension26:43
Learning curves for multivariate brain state decoding31:42
AIM II Interpretation: Visualization of networks34:13
Reproducibility of parameters/visualization? …hints from asymptotic theory37:24
The sensitivity map & the PR plot41:57
NPAIRS: Reproducibility of parameters43:12
Reproducibility of internal representations46:00
Unsupervised learning48:44
Unsupervised learning: Factor analysis generative model49:21
Factor models54:16
Matrix factorization: SVD/PCA, NMF, Clustering54:44
ICA: Assume S(k,t)’s statistically independent54:47
DTU:ICA toolbox56:05
More challenges for the linear factor model56:43
Beyond the linear model: Motivation59:10
De-noising by projection onto non-linear signal manifolds w/ kPCA01:01:13
bbci2012_hansen_neurotechnology_01_Page_3101:03:12
bbci2012_hansen_neurotechnology_01_Page_3201:04:11
bbci2012_hansen_neurotechnology_01_Page_3301:04:23
Beyond the linear model: Optimizing denoising using the PR-plot01:06:01
Does denoising by kernel PCA help fMRI decoding?01:06:34
Supervised learning01:06:56
Generalizable supervised models - ‘mind reading’01:07:26
Visualization of SVM learning from fMRI01:08:43
Visualization of kernel machine internal representations01:09:03
The sensitivity map01:09:09
Consistency across models (left-right finger tapping)01:09:46
Sensitivity maps for non-linear kernel regression01:11:27
Conclusions01:15:02
Outlook – future of mind reading?01:16:55
Acknowledgments01:20:54