1: \begin{abstract}

2: % In compressed sensing, the goal is to reconstruct the signal from a number of linear measurements. Since the problem in most cases is underdetermined, prior knowledge about the signal of interest is required. In this work, we propose an equivariant generative model as a prior for solving inverse problems. We show how to train the variational autoencoder with the equivariant latent space and provide additional theoretical recovery guarantees for generative priors. This equivariant VAE is then used as a prior in compressed sensing. We further show that this approach can be used to recover signals with unknown orientation. The proposed model brings additional potential gains in terms of convergence and latency. 

3: 

4: 

5: %%%%%%%%%%%%%%%%%%%%%%%%%%

6: %% AB Proposal

7: %%%%%%%%%%%%%%%%%%%%%%%%%%

8: 

9: In compressed sensing, the goal is to reconstruct the signal from an underdetermined system of linear measurements. Thus, prior knowledge about the signal of interest and its structure is required. Additionally, in many scenarios, the signal has an unknown orientation prior to measurements. To address such recovery problems, we propose using equivariant generative models as a prior, which encapsulate orientation information in their latent space. Thereby, we show that signals with unknown orientations can be recovered with iterative gradient descent on the latent space of these models and provide additional theoretical recovery guarantees. We construct an equivariant variational autoencoder and use the decoder as generative prior for compressed sensing. We discuss additional potential gains of the proposed approach in terms of convergence and latency. 

10: % We introduce Equivariant VAE model, which can be used as a generative prior for inverse linear problems. We show that this helps to improve quality of the reconstructions in case of rotations ambiguities of the input.

11: \end{abstract}

12: