Here we have the results of a dataset containing recordings from patients suffering with epilepsy. The dataset was used in a competition at Kaggle's website (link), where competitors had to classify different intracranial EEG recordings according to their distance to an epileptic seizure. In the following example we used several 1s clips along with their labels of "time to next seizure".
The figure below shows the results of an embedding using diffusion maps for one patient. Each point corresponds to a 36x36 covariance matrix and the colors indicate how far the corresponding short signal is from a seizure. The latencies go from 120 seconds (green) all the way down to 0 seconds (red) to seizure.
These are the results for the embedding of another patient. The points now correspond to 30x30 covariance matrices and the latencies go from 60 seconds to 0 seconds to seizure.