NullFlow does one-step generative reconstruction by keeping the entire flow inside a measurement-consistent subspace. Since the trajectory never leaves that subspace, no separate data-fidelity correction step is needed — unlike existing solvers that alternate between generation and projection.
@misc{shi2026NullFlow,title={NullFlow: One-Step Generative Reconstruction},author={Shi, Xiao and Chandler, Edward P. and Park, Chicago Y. and Chandler, Edward P. and Shoushtari, Shirin and Kamilov, Ulugbek S.},year={2026},}
Sparse strides capture global structure, and progressively denser strides add fine detail. DenseAR follows this principle, autoregressively predicting tokens in this coarse-to-fine order — on a single compact latent grid, with no multi-scale token inflation.
@misc{park2026DenseAR,title={Next-Dense-Stride Prediction for Multimodal Autoregressive Visual Modeling},author={Park, Chicago Y. and Mao, Jialin and Xu, Xiaojian and Kass-Hout, Taha and Kamilov, Ulugbek S. and Xiao, Cao},year={2026},}
PnP methods plug a denoiser into an optimization loop; diffusion models train a denoiser across all noise levels. This paper connects the two — showing a score-based reading of PnP that justifies dropping a pretrained diffusion model directly into a PnP algorithm, no reverse sampling needed.
@misc{park2026SGPnP,title={Stochastic Generative Plug-and-Play Priors},author={Park, Chicago Y. and Chandler, Edward P. and Hu, Yuyang and McCann, Michael T. and Garcia-Cardona, Cristina and Wohlberg, Brendt and Kamilov, Ulugbek S.},year={2026},}
We propose a simple, architecture-agnostic method for conditioning deep generative models on time or noise level by interpolating between two learnable parameter sets.
@misc{park2026deepparameterinterpolation,title={Deep Parameter Interpolation for Scalar Conditioning},author={Park, Chicago Y. and McCann, Michael T. and Garcia-Cardona, Cristina and Wohlberg, Brendt and Kamilov, Ulugbek S.},booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition Findings (CVPRF)},year={2026},}
We propose the first measurement score-based diffusion model that directly learns partial measurement scores using only noisy and subsampled measurements, enabling the synthesis of fully sampled measurements and solving inverse problems.
@misc{park2026measurementdiffusion,title={Measurement Score-Based Diffusion Model},author={Park, Chicago Y. and Shoushtari, Shirin and An, Hongyu and Kamilov, Ulugbek S.},booktitle={International Conference on Learning Representations (ICLR)},year={2026},}
We propose measurement score-based MRI reconstruction algorithms that do not require both ground-truth MRI images and precalibrated coil sensitivity maps.
@misc{liu2025measurementscoremri,title={Measurement Score-Based MRI Reconstruction with Automatic Coil Sensitivity Estimation},author={Liu, Tingjun and Park, Chicago Y. and Hu, Yuyang and An, Hongyu and Kamilov, Ulugbek S.},year={2025},}
We present a configurable theoretical framework for score-based diffusion models, leading to generic algorithm templates of influential diffusion models and enabling faster and more flexible sampling strategies.
@article{park2025randomwalks,title={Random Walks With Tweedie: A Unified View of Score-Based Diffusion Models [In the Spotlight]},author={Park, Chicago Y. and McCann, Michael T. and Garcia-Cardona, Cristina and Wohlberg, Brendt and Kamilov, Ulugbek S.},journal={IEEE Signal Processing Magazine},volume={42},number={3},pages={40--51},year={2025},publisher={IEEE},doi={10.1109/MSP.2025.3590608},url={https://doi.org/10.1109/MSP.2025.3590608}}
We reinterpret plug-and-play (PnP) methods as score-based approaches, enabling the use of score-based diffusion model (SBM) priors in PnP without retraining, allowing direct comparison with SBM-based reconstruction methods.
@inproceedings{park2025pnpscore,title={Plug-and-Play Priors as a Score-Based Method},author={Park, Chicago Y. and Hu, Yuyang and McCann, Michael T. and Garcia-Cardona, Cristina and Wohlberg, Brendt and Kamilov, Ulugbek S.},booktitle={IEEE International Conference on Image Processing (ICIP)},year={2025},}
We propose the new approach combining network pruning and fine-tuning to enhance the efficiency of model-based deep learning for imaging inverse problems.
@article{park2023pruning,title={Efficient Model-based Deep Learning via Network Pruning and Fine-Tuning},author={Park, Chicago Y. and Gan, Weijie and Zou, Zihao and Hu, Yuyang and Sun, Zhixin and Kamilov, Ulugbek S.},journal={Journal of Mathematical Imaging and Vision},year={2025},}
We propose the theoretical and experimental evidence for the effective use of expansive CNN denoisers in the PnP-ADMM framework to solve convex or non-convex imaging inverse problems.
@inproceedings{park2023pnpadmm,author={Park, Chicago. Y. and Shoushtari, S. and Gan, W. and Kamilov, Ulugbek S.},booktitle={IEEE International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP)},title={Convergence of Nonconvex PnP-ADMM with MMSE Denoisers},year={2023},address={Los Suenos, Costa Rica},pages={511--515},}