True -- the available data sets (like MIT-BIH arrhythmia database) are quite small. In our case, we launched a study with UCSF Cardiology which has recruited about 10,000 people so far, and that's how we get a data set large enough for deep learning.