Compressed Sensing of Generative Sparse-latent (GSL) Signals
This work addresses compressed sensing for signals with generative models, which is incremental as it builds on existing methods by incorporating sparse-latent structures.
The paper tackles the problem of reconstructing ambient signals in compressed sensing using a neural network generative model with sparse-latent inputs, showing that a gradient-based non-convex algorithm achieves good reconstruction performance in simulations.
We consider reconstruction of an ambient signal in a compressed sensing (CS) setup where the ambient signal has a neural network based generative model. The generative model has a sparse-latent input and we refer to the generated ambient signal as generative sparse-latent signal (GSL). The proposed sparsity inducing reconstruction algorithm is inherently non-convex, and we show that a gradient based search provides a good reconstruction performance. We evaluate our proposed algorithm using simulated data.