The Pseudo Projection Operator: Applications of Deep Learning to Projection Based Filtering in Non-Trivial Frequency Regimes
This work addresses filtering challenges in non-trivial frequency regimes for applications in physical and biological sciences, but it is incremental as it combines existing neural network techniques with projection operators.
The paper tackles the problem of separating signal from noise in frequency-based projection filters, which traditionally require prior knowledge of non-overlapping signal and noise frequencies, by introducing a hybrid model called the Pseudo Projection Operator (PPO) that uses a neural network for frequency selection. The result shows that the PPO outperforms traditional projection operators and denoising autoencoders in most experiments on a music performance dataset with various noise types.
Traditional frequency based projection filters, or projection operators (PO), separate signal and noise through a series of transformations which remove frequencies where noise is present. However, this technique relies on a priori knowledge of what frequencies contain signal and noise and that these frequencies do not overlap, which is difficult to achieve in practice. To address these issues, we introduce a PO-neural network hybrid model, the Pseudo Projection Operator (PPO), which leverages a neural network to perform frequency selection. We compare the filtering capabilities of a PPO, PO, and denoising autoencoder (DAE) on the University of Rochester Multi-Modal Music Performance Dataset with a variety of added noise types. In the majority of experiments, the PPO outperforms both the PO and DAE. Based upon these results, we suggest future application of the PPO to filtering problems in the physical and biological sciences.