Reviews: Random Projection Filter Bank for Time Series Data

This paper proposes a new algorithm for learning on time series data. The basic idea is to introduce random projection filter bank for extracting features by convolution with the time series data. Learning on time series data can then be performed by applying conventional machine learning algorithms to these features. Theoretical analysis is performed by studying both the approximation power and statistical guarantees of the introduced model. Experimental results are also presented to show its effectiveness in extracting features from time series.