Atrial fibrillation (AFib) is one of the most common sustained cardiac arrhythmia, and is associated with significant mortality and morbidity through association of risk of death, stroke, hospitalization, heart failure and coronary artery disease. AFib detection from single-lead ECG recordings is still an open problem, as AFib event may be episodic and the signal noisy. In this study we conduct a thoughtful analysis of recent deep network architectures developed in the computer vision field, redesigned to be suitable for a one dimensional signal, and we evaluate their performance for the AFib detection problem. We show evidence on the capabilities of data-driven automatic knowledge extraction to enhance the performance of biomedical signal analysis systems.