Cleansing Method of Ripple-firing Data based on Condorcet Jury Theorem

Ripple firings occurring in the hippocampal CA1 are considered in accordance with the memory of episodic experiences. Toward elucidating the relation, we developed a convolutional neural network (CNN) that classifies ripple firings into each category of the experienced episodes. However, its performance did not achieve a level of practical use. As a possible reason, we surmise that ripple firing data unrelated to each experienced episode was intermingled in the CNN training dataset. This letter proposes a method to cleanse such unrelated data from the dataset. The data to be cleansed is selected with majority votes in multinomial-logistic-regression-based weak classifiers, where the number of weak classifiers is determined by Condorcet jury theorem.