MVPA_research_diary
MVPA
Tasks and Ideas:
- Use shared-component GMM to model the location of the brain signal sources. This can be used in feature selection too.
- Try to use first k-component SVD to decompose the observation matrix. We don't need to represent the data in the beta space.
- Hierarchical Bayesian GMM
- Apply ICA to the individual beta map?
- Derive Gaussian EM for grid data based on Tori's paper
- Find the brain network in the task-oriented experiment, and compare the network with the activation areas obtained from the beta map!
- I'm running dimensionality reduction test on the the fMRI data..so far, MDS and LLE..still need to work on PCA, tSNE, SNE
- I will have to run some classifier on the data...SVM, LR, Decision Tree, Random Forest, AdaBoost
- Feature selection using Rand index, MI, whatever...
- Think about how to decompose the brain into components...GMM?
On-going work:
- Make similarity matrix for each pair of voxels using MI, then use spectral clustering to cluster them. We expect to see some cluster on the x y z coordinates
- In fact, we can use any kind of clustering, for instance GMM. It's a good idea to incorporate the x y z position too.
- Represent brain using r and theta coordinate?
- MI just talks about separability in all classes...what if we do separability for each class, say [c1 vs the rest], [c2 vs the rest], etc. This way we will know what voxel is the best to separate c1 from others, and we say that the voxel is corresponding for class c1. ---> we can plot the MI on the beta map
Readings:
Representational space
Distributed and Overlapping Representations of Faces and Objects in Ventral Temporal Cortex [url]
Classifier ensembles for fMRI data analysis: an experiment [url]
Beyond mind-reading: multi-voxel pattern analysis of fMRI data [url]
Predicting Human Brain Activity Associated with the Meanings of Nouns [url]
Circular analysis in systems neuroscience: the dangers of double dipping [url]
A Common, High-Dimensional Model of the Representational Space in Human Ventral Temporal Cortex [url]
Decoding the Representation of Multiple Simultaneous Objects in Human Occipitotemporal Cortex [url]
Aligning Brains and Minds [url]
A topographic latent source model for fMRI data [url] (NeuroImg 2011) <--**
A Bayesian Mixture Approach to Modeling Spatial Activation Patterns in Multisite fMRI Data [url] (IEEE Trans Med Img 2010) <--- *****
Modeling Inter-Subject Variability in fMRI Activation Location: A Bayesian Hierarchical Spatial Model [url] (Biometric 2009) <-- ***
The representation of biological classes in the human brain [url]
Searchlight approach
A comparison of volume-based and surface-based multi-voxel pattern analysis [url]
Slides or videos:
Overview of SPM: the SPM course video webpage from UCL
General Linear Model (GLM) applied to fMRI data analysis [pdf]
toolboxes:
Princeton's MVPA page
-------------------------------------------------------------------------------------------------
Regularizations for logistic regression — Feb 4, 2013 1:39:01 AM
Voxel selection using submodular function optimization — Feb 4, 2013 1:24:00 AM
Next steps for distinguishability measure — Oct 7, 2012 10:14:57 PM
Collaborative filtering on brain — Aug 24, 2012 4:34:24 PM
Conference/Journal list — Aug 21, 2012 6:09:25 PM
Voxel selection using l1-norm — Aug 15, 2012 7:49:13 PM
Justify the model and the data using earning curve analysis — Aug 15, 2012 12:56:56 AM
A story to tell — Jul 30, 2012 6:34:12 PM
I like Gael's website — Jul 23, 2012 8:38:13 PM
Surface-based searchlight approach — Jul 23, 2012 8:14:41 PM
Haxby et al.'s several-subject fMRI is available — Jun 15, 2012 2:52:42 AM
Using independent voxel classification accuracy to determine feature order — May 25, 2012 6:50:38 AM
Using supervoxels in fMRI classification — May 19, 2012 1:16:42 AM
GMM on the class specific response map — May 19, 2012 1:14:18 AM