User Reference:LinearClassifier - Revision history

Mellinger: /* Function */

2012-07-18T13:06:07Z

Function

← Older revision		Revision as of 13:06, 18 July 2012
Line 9:		Line 9:

	*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is later [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.		*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is later [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.
	*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent a log-likelihood ~~ratop~~ for each of these responses to be an ERP.		*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent a log-likelihood ratio for each of these responses to be an ERP.

	==Parameters==		==Parameters==

Mellinger: /* Function */

2011-08-23T14:11:58Z

Function

← Older revision		Revision as of 14:11, 23 August 2011
Line 9:		Line 9:

	*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is later [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.		*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is later [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.
	*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent ~~the~~ likelihood for each of these responses to be an ERP.		*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent a log-likelihood ratop for each of these responses to be an ERP.

	==Parameters==		==Parameters==

Mellinger: /* Function */

2009-10-19T08:31:12Z

Function

← Older revision		Revision as of 08:31, 19 October 2009
Line 4:		Line 4:
	has a single index (C channels x 1 element), thus the linear classifier		has a single index (C channels x 1 element), thus the linear classifier
	acts as a N x M x C matrix, determining the output after summation over		acts as a N x M x C matrix, determining the output after summation over
	N and M:		channels and elements:

	<math>\textrm{output}_k=\sum_{i=1}^N \sum_{j=1}^M \textrm{input}_{ij}\, \textrm{Classifier}_{ijk}</math>		<math>\textrm{output}_k=\sum_{i=1}^N \sum_{j=1}^M \textrm{input}_{ij}\, \textrm{Classifier}_{ijk}</math>

Mellinger: /* Function */

2009-06-05T08:23:40Z

Function

← Older revision		Revision as of 08:23, 5 June 2009
Line 8:		Line 8:
	<math>\textrm{output}_k=\sum_{i=1}^N \sum_{j=1}^M \textrm{input}_{ij}\, \textrm{Classifier}_{ijk}</math>		<math>\textrm{output}_k=\sum_{i=1}^N \sum_{j=1}^M \textrm{input}_{ij}\, \textrm{Classifier}_{ijk}</math>

	*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.		*In a BCI based on demodulation of a periodic EEG component (such as the [[User Tutorial:Introduction to the Mu Rhythm#BCI Construction\|Mu Rhythm BCI]]), the LinearClassifier's input is the time-varying amplitude or power spectrum for a number of spatially filtered EEG channels. Its output is later [[User Reference:Normalizer\|normalized with respect to mean and variance]], and then used as a [[BCI2000 Glossary#Control Signal\|control signal]] to determine cursor movement.
	*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent the likelihood for each of these responses to be an ERP.		*In an ERP-based BCI (such as the [[User Tutorial:Introduction to the P300 Response#BCI Construction\|P300 BCI]]), the LinearClassifier's input is a sequence of averaged EEG time courses obtained in response to a number of stimuli, and its output is considered to represent the likelihood for each of these responses to be an ERP.

Mellinger: /* 2D cursor movement */

2008-11-19T12:07:56Z

2D cursor movement

← Older revision		Revision as of 12:07, 19 November 2008
Line 56:		Line 56:
	\|}		\|}

	Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|~~Spatial Filter~~]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing spectral amplitudes involves taking the absolute value (or its square), which is a nonlinear operation, with <math>\|A-B\| \neq \|A\|-\|B\|</math> in general.		Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|SpatialFilter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing spectral amplitudes involves taking the absolute value (or its square), which is a nonlinear operation, with <math>\|A-B\| \neq \|A\|-\|B\|</math> in general.

	===ERP classification===		===ERP classification===

Mellinger: /* 2D cursor movement */

2008-11-19T12:06:36Z

2D cursor movement

← Older revision		Revision as of 12:06, 19 November 2008
Line 56:		Line 56:
	\|}		\|}

	Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing ~~the spectrum~~ involves taking the absolute value (or its square), which is a nonlinear operation, with <math>\|A-B\| \neq \|A\|-\|B\|</math> in general.		Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing spectral amplitudes involves taking the absolute value (or its square), which is a nonlinear operation, with <math>\|A-B\| \neq \|A\|-\|B\|</math> in general.

	===ERP classification===		===ERP classification===

Mellinger: /* 2D cursor movement */

2008-11-19T12:04:16Z

2D cursor movement

← Older revision		Revision as of 12:04, 19 November 2008
Line 56:		Line 56:
	\|}		\|}

	Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation:		Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation, with <math>\|A-B\| \neq \|A\|-\|B\|</math> in general.
	<math>\|A-B\| \neq \|A\|-\|B\|</math>

	===ERP classification===		===ERP classification===

Mellinger: /* 2D cursor movement */

2008-11-19T12:03:03Z

2D cursor movement

← Older revision		Revision as of 12:03, 19 November 2008
Line 51:		Line 51:
	\| CP4 \|\| 12Hz \|\| 1 \|\| 0.5		\| CP4 \|\| 12Hz \|\| 1 \|\| 0.5
	\|-		\|-
	\| CP3 \|\| 24Hz \|\| 2 \|\| 0.5		\| CP3 \|\| 24Hz \|\| 2 \|\| -0.5
	\|-		\|-
	\| CP4 \|\| 24Hz \|\| 2 \|\| -0.5		\| CP4 \|\| 24Hz \|\| 2 \|\| 0.5
	\|}		\|}

	Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, ~~the spectrum is computed between~~ spatial filtering and feature combination. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation:		Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, spatial filtering is followed with computing spectral amplitudes, and feature combination amounts to an addition of spectral amplitudes. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation:
	<math>\|A-B\| \neq \|A\|-\|B\|</math>		<math>\|A-B\| \neq \|A\|-\|B\|</math>

Mellinger at 16:37, 25 February 2008

2008-02-25T16:37:16Z

← Older revision		Revision as of 16:37, 25 February 2008
Line 28:		Line 28:

	==Examples==		==Examples==
	===~~SMR~~ Examples===		===Mu Rhythm Examples===
	For these examples, let us assume that you set the [[User Reference:ARFilter\|ARFilter]]'s ''FirstBinCenter'' to 0, and ''BinWidth'' to 3Hz. This allows you to refer to the respective bins by their frequency, i.e. ''12Hz'' rather than ''5''.		For these examples, let us assume that you set the [[User Reference:ARFilter\|ARFilter]]'s ''FirstBinCenter'' to 0, and ''BinWidth'' to 3Hz. This allows you to refer to the respective bins by their frequency, i.e. ''12Hz'' rather than ''5''.
	We also assume that, in the [[User Reference:SpatialFilter\|SpatialFilter]], you have entered labels for output channels. This allows you to refer to channels by their names, i.e. ''CP3'' rather than ''7''.		We also assume that, in the [[User Reference:SpatialFilter\|SpatialFilter]], you have entered labels for output channels. This allows you to refer to channels by their names, i.e. ''CP3'' rather than ''7''.

Mellinger: /* 2D cursor movement */

2008-02-25T16:36:48Z

2D cursor movement

← Older revision		Revision as of 16:36, 25 February 2008
Line 56:		Line 56:
	\|}		\|}

	Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for ~~an SMR~~ BCI, the spectrum is computed between spatial filtering and feature combination. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation:		Note that, in the above example, subtracting CP3 from CP4 in the classifier is not equivalent to taking the difference in the [[User Reference:SpatialFilter\|Spatial Filter]]. This is because, for a Mu Rhythm BCI, the spectrum is computed between spatial filtering and feature combination. Computing the spectrum involves taking the absolute value (or its square), which is a nonlinear operation:
	<math>\|A-B\| \neq \|A\|-\|B\|</math>		<math>\|A-B\| \neq \|A\|-\|B\|</math>