SPS Webinar: Directly Parameterized Neural Network Construction for Generalization and Robustness in Imaging Inverse Problems
Date: 17 July 2025
Time: 11:30 AM ET (New York Time)
Presenter(s): Dr. Nikola Janjušević, Dr. Amirhossein Khalilian-Gourtani
Based on the IEEE Xplore® article:
"CDLNet: Noise-Adaptive Convolutional Dictionary Learning Network for Blind Denoising and Demosaicing"
Published: IEEE Open Journal of Signal Processing, May 2022.
Download article: Original article is open access and publicly available for download.
Abstract
Deep learning models achieve state-of-the-art performance in image restoration but remain largely black-box, making it difficult to understand the role of their weights or adapt them to new settings. Algorithm unrolling offers a more principled mechanism to neural network architecture design. In this talk, the presenters will introduce a specific approach to algorithm unrolling via direct parameterization, the Convolutional Dictionary Learning Network (CDLNet). In their approach, they unroll a classical convolutional dictionary learning algorithm into a deep network, ensuring that each layer's operators have a direct correspondence to the original algorithm. They will discuss how this structured design allows CDLNet to achieve competitive results in image denoising and joint denoising-demosaicing while requiring fewer learned parameters than fully convolutional black-box models. Moreover, CDLNet’s interpretability enables the introduction of a noise-adaptive thresholding mechanism, leading to state-of-the-art blind denoising and near-perfect generalization to unseen noise levels. Beyond CDLNet, they will briefly explore how similar interpretability principles extend to architectures incorporating nonlocal self-similarity priors for denoising and compressed sensing MRI reconstruction. By replacing black-box deep models with structured, optimization-inspired networks, these approaches enhance both robustness and generalization, offering a promising path forward for deep learning in inverse problems.
Biography
Nikola Janjušević received the B.Eng. degree in electrical engineering with a minor in computer science from The Cooper Union, New York, NY, USA and the Ph.D. in electrical engineering from the New York University Tandon School of Engineering, Brooklyn, NY, USA in 2019 and 2024 respectively.
He is currently a postdoctoral research fellow at the New York University Grossman School of Medicine, Department of Radiology, New York, NY, USA. Previously, he was a research scientist at Apple Video Engineering and Samsung Research America. His research interests lie at the intersection of imaging inverse problems, interpretable deep learning, non-smooth and convex optimization, and medical imaging. He has published journal papers on interpretable noise-adaptive deep learning architectures and compressed sensing optical coherence tomography.
Amirhossein Khalilian-Gourtani received the B.Sc. degree in electrical engineering and communications from the Isfahan University of Technology, Isfahan, Iran, the M.Sc. degree in electrical engineering from New York University, New York, NY, USA, in 2018 and the Ph.D. in electrical engineering from the New York University Tandon School of Engineering in 2022.
He is currently a postdoctoral research fellow at the New York University Grossman School of Medicine, Department of Neurology. His main research interests include signal processing, machine learning, convex and non-smooth optimization, medical signal processing, and numerical analysis.
Dr. Khalilian-Gourtani was the recipient of Postdoctoral Merit Award from Society for the Neurobiology of Language, the Ernst Weber PhD Fellowship Award, and the Myron M. Rosenthal M.Sc. Award from New York University. He is a reviewer of more than ten computer vision related journals and conferences of IEEE.
