Neural network layer code written using Keras to implement Wavelet Deconvolutions from the paper:
Khan, Haidar, and Bulent Yener. "Learning filter widths of spectral decompositions with wavelets." Advances in Neural Information Processing Systems. 2018.
Requires Keras with a Tensorflow backend in addition to standard packages such as numpy, matplotlib, scipy, and h5py.
Run testWD.py to verify model saving, model loading, and proper functionality.
Deconvolutions of 1D signals using wavelets of different scales/widths. For a full description of the wavelet deconvolution method, see our paper
# apply a set of 5 wavelet deconv widthss to a sequence of 32 vectors with 10 timesteps
model = Sequential()
model.add(WaveletDeconvolution(5, kernel_length=200, padding='same', input_shape=(32, 10), data_format='channels_first'))
# now model.output_shape == (None, 32, 10, 5)
# add a conv2d on top
model.add(Convolution2D(64, 3, 3, padding='same'))
# now model.output_shape == (None, 64, 10, 5)