Data scaling in remote health monitoring systems

We formalize the data scaling problem as the ability to scale down computations of stream analysis software components. Data scaling enables systems to trade computational accuracy for resources. We develop an information theoretic technique to classification problems in remote health monitoring and propose two methods for trading computational utility for bandwidth. Experiments on ECG classification reveal the potential of this approach by reporting significant resource savings for small amounts of utility degradation, e.g., 33% of bandwidth saving for only a 1% of accuracy degradation.