Problems of representation of electrocardiograms in convolutional neural networks
This paper highlights fundamental limitations of CNNs for researchers and practitioners working with time-series data where patterns exhibit variability, suggesting that current approaches may be suboptimal.
This paper identifies systemic problems when standard convolutional neural networks are used to model one-dimensional signals, such as electrocardiograms, that contain inaccurate repeating patterns. The issues stem from the networks' handling of composite objects with significant part mobility, leading to counterintuitive generalization effects.
Using electrocardiograms as an example, we demonstrate the characteristic problems that arise when modeling one-dimensional signals containing inaccurate repeating pattern by means of standard convolutional networks. We show that these problems are systemic in nature. They are due to how convolutional networks work with composite objects, parts of which are not fixed rigidly, but have significant mobility. We also demonstrate some counterintuitive effects related to generalization in deep networks.