Gaussian process robust regression for noisy heart rate data

We obtained ECG recordings from the MIT Long-term database at www.physionet.org and played them back to the heart rate sensor (Actiheart) via a digital-to-analog converter. Subsequently, we used our proposed method to regress the data recorded by the device. Comparison between these results, the baseline model results, and the calculated true heart rate based on the beat-by-beat expert annotations provide further
evidence that our scheme represents a significant improvement over existing postprocessing methodology and the raw heart rate of the device.