Machine Learning Models to Automatically Discover Novel Functional Patterns in Multivariate Time Series

In this paper, we propose a method and a specific architecture for a machine learning model that assists researchers across various fields to automatically identify functional patterns in multivariate time series from a series of experiments. The initial problem is formalized in terms of machine learning, eliminating the need for researchers to be experts in the specific subject matter under examination. The effectiveness of the method is demonstrated in the field of neurophysiology with data where the existence of the P300 pattern is already known. For further research, it would be beneficial to generalize the proposed method to other areas, such as sensor data from production lines or banking transactions.