SGD-Net: Efficient Model-Based Deep Learning with Theoretical Guarantees
This work addresses scalability issues for researchers and practitioners in computational imaging, though it is incremental as it builds on existing deep unfolding methods.
The paper tackles the computational and memory inefficiency of deep unfolding networks in large-scale imaging inverse problems by proposing SGD-Net, which uses stochastic approximations to reduce complexity while matching batch network performance, as shown in tests on intensity diffraction tomography and sparse-view computed tomography.
Deep unfolding networks have recently gained popularity in the context of solving imaging inverse problems. However, the computational and memory complexity of data-consistency layers within traditional deep unfolding networks scales with the number of measurements, limiting their applicability to large-scale imaging inverse problems. We propose SGD-Net as a new methodology for improving the efficiency of deep unfolding through stochastic approximations of the data-consistency layers. Our theoretical analysis shows that SGD-Net can be trained to approximate batch deep unfolding networks to an arbitrary precision. Our numerical results on intensity diffraction tomography and sparse-view computed tomography show that SGD-Net can match the performance of the batch network at a fraction of training and testing complexity.