Summary

- Plugin name : Independent Component Analysis (FastICA)
- Version : 0.2
- Author : Guillaume Gibert / Jeff B.
- Company : INSERM / Independent
- Short description : Computes fast independent component analysis
- Documentation template generation date : Dec 30 2016
- WARNING : this box has been marked as UNSTABLE by the developer. It means that its implementation may be incomplete or that the box can only work under well known conditions. It may possibly crash or cause data loss. Use this box at your own risk, you've been warned.
Description
This box attempts to find a decomposition of the signal to its independent components. The approach is based on the FastICA algorithm.
Inputs
1. Input signal
The input signal.
- Type identifier : Signal (0x5ba36127, 0x195feae1)
Outputs
1. Output signal
The decomposed signal.
- Type identifier : Signal (0x5ba36127, 0x195feae1)
Settings
1. Number of components to extract
Number of independent components to extract (equals PCA dimension reduction)
- Type identifier : Integer (0x007deef9, 0x2f3e95c6)
- Default value : [ 4 ]
2. Operating mode
Which decomposition is desired?
- Type identifier : Operating mode (0x43a71032, 0x4af96b9f)
- Default value : [ ICA ]
3. Sample size (seconds) for estimation
How many seconds of sample to collect to estimate the ICA model?
- Type identifier : Integer (0x007deef9, 0x2f3e95c6)
- Default value : [ 120 ]
4. Decomposition type
Decomposition type. Deflation is an approach where each component is estimated separately in turns. Symmetric estimation optimizes all components at once.
- Type identifier : Decomposition type (0x7b876033, 0x13590b93)
- Default value : [ Symmetric ]
5. Max number of reps for the ICA convergence
Maximum number of iterations
- Type identifier : Integer (0x007deef9, 0x2f3e95c6)
- Default value : [ 100000 ]
6. Fine tuning
Enable fine tuning?
- Type identifier : Boolean (0x2cdb2f0b, 0x12f231ea)
- Default value : [ true ]
7. Max number of reps for the fine tuning
Maximum number of iterations for the fine tuning
- Type identifier : Integer (0x007deef9, 0x2f3e95c6)
- Default value : [ 100 ]
8. Used nonlinearity
Used nonlinearity type
- Type identifier : Nonlinearity (0x4313472f, 0x37fd5961)
- Default value : [ Tanh ]
9. Internal Mu parameter for FastICA
Mu parameter
- Type identifier : Float (0x512a166f, 0x5c3ef83f)
- Default value : [ 1.0 ]
10. Internal Epsilon parameter for FastICA
Epsilon parameter
- Type identifier : Float (0x512a166f, 0x5c3ef83f)
- Default value : [ 0.0001 ]
11. Spatial filter filename
Filename to save the estimated decomposition matrix W to
- Type identifier : Filename (0x330306dd, 0x74a95f98)
- Default value : [ ]
12. Save the spatial filter/demixing matrix
Should the matrix W be saved to a file?
- Type identifier : Boolean (0x2cdb2f0b, 0x12f231ea)
- Default value : [ false ]
Examples
One use-case of ICA is to attempt to separate the signal of interest from nuisance artifacts. For example, supposing that ICA makes a meaningful decomposition of your EEG signal, you will see artifacts such as those from eyeblinks more clearly segregated to specific output channels instead of contaminating all of the channels.
Miscellaneous
This plugin applies the FastICA algorithm to the input signal. The box can store the estimated decomposition matrix W to a file. This file can then be used later in the spatial filter box to apply the decomposition on fresh data.
The box also outputs the decomposed signal, but the decomposition is active only after the model has been estimated (after the specified number of samples have been collected). If you wish to decompose the whole data, then you can first train the ICA model, save the matrix, and then separately apply it to the original data with the spatial filter.
The FastICA algorithm is described in
A. Hyvärinen. "Fast and Robust Fixed-Point Algorithms for Independent Component Analysis", IEEE Transactions on Neural Networks 10(3):626-634, 1999.
The implementation used by the box is from the ITPP toolkit.
Generated on Tue Jun 26 2012 15:25:54 for Documentation by
