Applying deep learning to single-trial EEG data provides evidence for complementary theories on action control.
Amirali Vahid1, Moritz Mückschel1, Sebastian Stober2
1Cognitive Neurophysiology, Department of Child and Adolescent Psychiatry, Faculty of Medicine, TU Dresden, Germany.
Communications Biology
|March 11, 2020
Summary
Deep learning accurately predicts action conflicts using single-trial electroencephalography (EEG) data. This approach identifies neural processes related to attention and response selection, advancing cognitive neuroscience.
Area of Science:
- Cognitive Neuroscience
- Artificial Intelligence
- Neurophysiology
Background:
- Action control is crucial for goal-directed behavior.
- Theories emphasize attention or response selection in action control.
- Identifying these processes in single-trial neural dynamics remains challenging.
Purpose of the Study:
- To determine if single-trial electroencephalography (EEG) data can predict action conflicts.
- To investigate the role of attention and response selection in predicting conflicts.
- To explore the utility of deep learning in linking cognitive theory and neurophysiology.
Main Methods:
- Applied deep learning algorithms to single-trial EEG data.
- The deep learning model was not pre-informed by cognitive theories.
- Analyzed neurophysiological features contributing to prediction accuracy.
Main Results:
- Deep learning predicted conflict presence in ~95% of subjects, significantly above chance level (33%).
- Key predictive features originated from attentional and motor response selection processes.
- Involved neural activity in the occipital cortex and superior frontal gyrus.
Conclusions:
- Deep learning can effectively identify predictive neurophysiological processes in single-trial neural dynamics.
- Artificial intelligence approaches can validate and develop links between cognitive theory and neurophysiology.
- This study demonstrates a powerful method for understanding the neural basis of action control.
