Sumários
Linear Discriminat Analysis 3 (online class)
28 maio 2020, 12:00 • Agostinho Cláudio da Rosa
Confusion Matrix
Independent Component Analysis (ICA) (online class)
26 maio 2020, 14:00 • Agostinho Cláudio da Rosa
Tools and functions to perform the lab are readily available in Matlab or provided under current lab
directory.
1 – MEMD and (Fast)ICA applications:
a. 5% Read the sleep signals from the edf format file lab09.edf. Form a data matrix X with 9
signals starting at 10 until 30 seconds (i.e. 20 seconds duration) of channels E1-M2, E2-M1, F3-M2, C3-
M2, O1-M2, F4-M1, C4-M1, O2-M1 and X2). Normalize X and plot all channels in one figure with 2
columns.
b. 25% Perform a MEMD analysis of matrix X (tictoc). From the resulting IMFs identify the
maximum number of frequency bands components , transients Fast and changes), sources (ECG, EOG,..),
artifacts (movement, muscle) and background noise(power line).
c. 20% For all signals plot the spectrum of its IMF and calculate in a table the area of the
spectral crossover between consecutive IMFs.
d. 25% Perform an ICA decomposition of X. Remove the EOG present. Plot the ICA
decompositions. Plot the EOG artifacts filtered decompositions. Provide relevant comments about the
quality of the filtered signals.
e. 10% Remove any artifacts and background noise, describe the process with detail. Comment
about the possibility to remove trend artifact.
f. 10% Compare the approaches in b., c. and e. in relation to artifacts removal.
g. 5% In an ICA problem, is it possible in general to tell the number of independent EEG
sources? Justify
h. 20% (Optional) Extract the ECG signal using the method used in d. compute the time and
frequency domain heart rate variability (HRV) parameters from the R-R Interval (RRI) curve. (Note: The
RRI curve have a not uniform sampling and very limited number of samples. Suggestion: use the Hertz
Equivalent resampling prior to frequency analysis)
Independent Component Analysis (ICA) (online class)
26 maio 2020, 12:30 • Agostinho Cláudio da Rosa
Tools and functions to perform the lab are readily available in Matlab or provided under current lab
directory.
1 – MEMD and (Fast)ICA applications:
a. 5% Read the sleep signals from the edf format file lab09.edf. Form a data matrix X with 9
signals starting at 10 until 30 seconds (i.e. 20 seconds duration) of channels E1-M2, E2-M1, F3-M2, C3-
M2, O1-M2, F4-M1, C4-M1, O2-M1 and X2). Normalize X and plot all channels in one figure with 2
columns.
b. 25% Perform a MEMD analysis of matrix X (tictoc). From the resulting IMFs identify the
maximum number of frequency bands components , transients Fast and changes), sources (ECG, EOG,..),
artifacts (movement, muscle) and background noise(power line).
c. 20% For all signals plot the spectrum of its IMF and calculate in a table the area of the
spectral crossover between consecutive IMFs.
d. 25% Perform an ICA decomposition of X. Remove the EOG present. Plot the ICA
decompositions. Plot the EOG artifacts filtered decompositions. Provide relevant comments about the
quality of the filtered signals.
e. 10% Remove any artifacts and background noise, describe the process with detail. Comment
about the possibility to remove trend artifact.
f. 10% Compare the approaches in b., c. and e. in relation to artifacts removal.
g. 5% In an ICA problem, is it possible in general to tell the number of independent EEG
sources? Justify
h. 20% (Optional) Extract the ECG signal using the method used in d. compute the time and
frequency domain heart rate variability (HRV) parameters from the R-R Interval (RRI) curve. (Note: The
RRI curve have a not uniform sampling and very limited number of samples. Suggestion: use the Hertz
Equivalent resampling prior to frequency analysis)
Linear Discriminant Analysis 2
26 maio 2020, 09:30 • Agostinho Cláudio da Rosa
2 Classes Numerical Example
Caannonical Correlation Analysis - CCA (online class)
21 maio 2020, 12:00 • Agostinho Cláudio da Rosa
projection maximizing the cross correlation