The paper presents a unified framework for image reconstruction called AUTOMap (Automated Transform by Manifold Approximation), which leverages deep neural networks to learn optimal reconstruction transforms for various imaging modalities. The framework aims to address the challenges of traditional image reconstruction methods, which often rely on ad hoc stages and expert parameter tuning. AUTOMAP is designed to be a data-driven, supervised learning task that maps sensor domain data to image domain output, implicitly learning a low-dimensional joint manifold that captures robust features of the data. The authors demonstrate the effectiveness of AUTOMAP through experiments on MRI acquisitions, showing superior noise immunity and artifact reduction compared to conventional methods. The framework is flexible and can be applied to a wide range of imaging modalities, including MRI, CT, PET, and radio astronomy. The paper also discusses the theoretical underpinnings of AUTOMAP, including the learning of stochastic projection operators and denoising steps, and provides detailed experimental results and analyses.The paper presents a unified framework for image reconstruction called AUTOMap (Automated Transform by Manifold Approximation), which leverages deep neural networks to learn optimal reconstruction transforms for various imaging modalities. The framework aims to address the challenges of traditional image reconstruction methods, which often rely on ad hoc stages and expert parameter tuning. AUTOMAP is designed to be a data-driven, supervised learning task that maps sensor domain data to image domain output, implicitly learning a low-dimensional joint manifold that captures robust features of the data. The authors demonstrate the effectiveness of AUTOMAP through experiments on MRI acquisitions, showing superior noise immunity and artifact reduction compared to conventional methods. The framework is flexible and can be applied to a wide range of imaging modalities, including MRI, CT, PET, and radio astronomy. The paper also discusses the theoretical underpinnings of AUTOMAP, including the learning of stochastic projection operators and denoising steps, and provides detailed experimental results and analyses.