Research
Our work follows three connected directions. Each is listed below with related papers.
Multimodal Perception and Decision Systems
Multimodal models that combine images, video, audio, and text, and decide when the visual evidence is sufficient to act. Robots look again before acting, and clinical models defer when uncertain.
- Vision-language models and multimodal perception
- Reliable multimodal reasoning under uncertainty in adaptive decision-making
- Human-centered perception and decision support
- Vision-language models for robot perception under partial observability
Related papers
- Prediction of Response to Spinal Cord Stimulation Using Machine Learning Based on Radiomics and Patient-Reported OutcomesNeurosurgery, 2026
- Multimodal Learning for Enhanced Detection in Oral Cancer ScreeningIEEE International Symposium on Biomedical Imaging (ISBI), 2025
Medical Image and Video Understanding
Ophthalmic imaging, surgical video, human movement, and brain connectivity. We interpret retinal scans, estimate camera motion inside the body, track human pose over time, and build tools that assist clinicians during a procedure.
- Foundation and self-supervised models for medical image interpretation
- Ophthalmic AI for retinal imaging and clinical decision support
- Surgical AI for workflow understanding and intelligent assistance
- Human pose estimation and motion analysis from video
- Graph learning on brain connectivity
Related papers
- SegSTRONG-C: Segmenting Surgical Tools Robustly On Non-adversarial Generated 'Corruptions', an EndoVis'24 ChallengeIEEE Transactions on Medical Imaging, 2026
- RVO-MIS: Robust Visual Odometry for Minimally Invasive SurgeryMedical Imaging with Deep Learning (MIDL), 2026
- Self-Supervised Graph Transformer with Contrastive Learning for Brain Connectivity Analysis towards Improving Autism DetectionIEEE International Symposium on Biomedical Imaging (ISBI), 2025
Reliable and Efficient AI Systems
Models that hold up under distribution shift and still run on the hardware available at the bedside or in the field. We study test-time adaptation and knowledge distillation toward this goal.
- Reliability under distribution and domain shift, including test-time adaptation
- Knowledge distillation and compact models for resource-constrained deployment
- Efficient adaptation of foundation models, with simulation and synthetic data
Related papers
- Efficient Hyperspectral Image Classification Using Discrete Cosine Transform on Limited-Resource SystemsIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025
- BCG-Former: Toward Pareto-Efficient Hyperspectral Image Classification via Band-Contextual GatingIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026
Collaborations
Our clinical and academic partners include:
- Department of Ophthalmology and Vision Science, University of Arizona
- Division of Cardiothoracic Surgery and Department of Emergency Medicine, University of Arizona College of Medicine, Tucson
- Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School