MODELING THE RELATIONSHIP BETWEEN ACOUSTIC STIMULUS AND EEG WITH A DILATED CONVOLUTIONAL NEURAL NETWORK (EUSIPCO2020)

Authors

Bernd Accou, Mohammad Jalilpour Monesi, Jair Montoya Martinez, Hugo Van hamme, Tom Francart

Models

Code for the baseline decoder model, baseline convolutional model and the dilated convolutional network can be found in baseline_decoder_model.py, baseline_convolutional_model.py and dilated_model.py respectively.

Models are constructed using the Keras API of tensorflow and can be trained by calling the fit method:

Linear decoder baseline model

Training the linear baseline is equivalent to training the linear decoder

from baseline_model import linear_decoder, linear_decoder_baseline

decoder = linear_decoder()
decoder.fit(eeg_train, good_env_train)

baseline = linear_decoder_baseline(decoder)
results = baseline.evaluate([eeg_test, env1_test, env2_test])

Convolutional baseline model

Training/Evaluating the simple convolutional model is as described in the Keras documentation

from baseline_convolutional_model import simple_convolutional_model

conv_model = simple_convolutional_model()
conv_model.fit([eeg_train, env1_train, env2_train], labels)

results = conv_model.evaluate([eeg_test, env1_test, env2_test], labels)

Dilated model

Training/Evaluating the dilated model is as described in the Keras documentation

from dilated_model import dilated_model

dilated = dilated_model()
dilated.fit([eeg_train, env1_train, env2_train], labels)

results = dilated.evaluate([eeg_test, env1_test, env2_test], labels)