Continuous Generative Neural Networks: A Wavelet-Based Architecture in Function Spaces
This work addresses inverse problems in infinite-dimensional settings, such as signal processing, by providing a theoretical framework for generative models, though it is incremental as it builds on existing architectures like DCGAN.
The authors tackled the problem of generative modeling in infinite-dimensional function spaces by introducing Continuous Generative Neural Networks (CGNNs), achieving injectivity conditions and Lipschitz stability for inverse problems, with numerical validation in tasks like signal deblurring.
In this work, we present and study Continuous Generative Neural Networks (CGNNs), namely, generative models in the continuous setting: the output of a CGNN belongs to an infinite-dimensional function space. The architecture is inspired by DCGAN, with one fully connected layer, several convolutional layers and nonlinear activation functions. In the continuous $L^2$ setting, the dimensions of the spaces of each layer are replaced by the scales of a multiresolution analysis of a compactly supported wavelet. We present conditions on the convolutional filters and on the nonlinearity that guarantee that a CGNN is injective. This theory finds applications to inverse problems, and allows for deriving Lipschitz stability estimates for (possibly nonlinear) infinite-dimensional inverse problems with unknowns belonging to the manifold generated by a CGNN. Several numerical simulations, including signal deblurring, illustrate and validate this approach.