Trainable Wavelet Neural Network for Non-Stationary Signals
Created on 2023-02-28T06:40:12-06:00
Jason Stock, Chuck Anderson. 2022
https://doi.org/10.48550/arXiv.2205.03355
Refers to the techique as "WaveNet," though that name probably has already been trounced by WaveRNN and co.
Classifies cloud and gravity waves.
- Performs wavelet analysis to produce coefficients as parameters for the network
Was useful after only ~10 epochs; though it could be trained longer its efficacy oscillates.
Only the wavelet is used--no convolution kernels with CNN nodes were used.