Deep Unrolling

Deep unrolling is a technique that transforms iterative optimization algorithms into deep neural networks, aiming to leverage the strengths of both model-based and purely data-driven approaches. Current research focuses on applying this method to diverse inverse problems, including image reconstruction (e.g., computed tomography, ptychography) and anomaly detection, often employing architectures based on quasi-Newton methods, stochastic gradient descent, or Bayesian frameworks. This approach offers improved efficiency and accuracy compared to traditional methods, particularly for large-scale problems, while also enhancing interpretability and enabling uncertainty quantification in critical applications like medical imaging and network security.