My work explores image generation, federated learning, and AI systems for healthcare. I am interested in building learning methods that remain useful when data is distributed, privacy-sensitive, noisy, or limited.
Recent projects include federated visual primitive sharing, reliable multi-label medical image diagnosis, LLM-orchestrated CT reconstruction, and deep learning methods for medical imaging tasks such as OCT speckle reduction, CT reconstruction, low-dose CT denoising, and MRI reconstruction.
Venue abbreviations follow each entry in brackets; [TL;DR] expands a short summary and titles link to the paper. *equal contribution, †corresponding author.
Accepted to IJCAI 2026. The work explores an LLM-orchestrated diagnose-plan-treat framework for mixed-degradation CT reconstruction.
Accepted as an ICML 2026 regular paper. The work studies reliable multi-label medical image diagnosis through fuzzy alignment with comorbidity topology.
Studies federated visual primitive sharing with text-guided adaptation, aiming to improve visual learning beyond fixed class boundaries in distributed settings.
Develops a trustworthy disentangled framework for multi-label medical image classification with multimodal refinement.
Proposes a self-supervised CT metal artifact reduction method based on range-null space decomposition and implicit neural representation.
Introduces an uncertainty-driven hierarchical sampling and retrieval framework for continual malware detection under class imbalance and concept drift.
Reviews federated learning for large models across the medical imaging pipeline, from reconstruction to clinical diagnosis and segmentation.
Rethinks how information is represented and exchanged in federated learning, aiming to improve collaborative training under distributed data settings.
Proposes a multi-scale and multi-level neural architecture search strategy for memory-efficient low-dose CT denoising.
Develops an attention-LSTM based multi-task method for short-term reactive and active load forecasting.
Applies neural architecture search to low-dose CT denoising, seeking effective model structures for medical image restoration.
Proposes a manifold and graph integrative convolutional network for low-dose CT reconstruction.
Develops a parameter-dependent framework for CT reconstruction across multiple geometries and dose levels.
Presents a sequential multi-task joint learning network for multiple stages of the MR imaging pipeline.
Introduces a dual-domain adaptive-scaling non-local network for reducing metal artifacts in CT imaging.
Presents a dual-domain adaptive-scaling non-local network for reducing metal artifacts in computed tomography.
Proposes multi-scale and multi-level neural architecture search for low-dose CT denoising.
Proposes a noise-powered disentangled representation method for unsupervised speckle reduction in optical coherence tomography images.
Introduces an unsupervised disentanglement network for speckle reduction in optical coherence tomography images.
Presents a parallel dual-domain convolutional neural network for compressed sensing MRI reconstruction.
Uses a generative adversarial network to perform simultaneous denoising and super-resolution for optical coherence tomography images.