|
Research
Continual Learning: learn new knowledge incrementally while not forgetting the old knowledge.
Online continual learning: Learn new knowledge incrementally from data stream where each data is observed only once by the model
Unsupervised/Self-supervised continual learning: Learn new knowledge or feature representation incrementally from unlabeled data
Continual learning in real world applications: Apply continual learning in various real-world applications such as food recognition
Image-based Dietary Assessment: Develop robust algorithms to automate the process of dietary assessment by using eating occasion images
Food recognition: Identify food type from eating occasion images in various scenarios and environments
Food portion size estimation: Estimate food energy/volume directly from eating occasion images
Food image generation: Generate synthetic and diverse food images using generative models
3D food data: Collect 3D food model and apply it for various 3D related downstream tasks
Long-tailed Learning: Learn knowledge from severe class-imbalanced data
Long-tailed classification: Learn unbiased feature representation and classifier for image classification
Long-tailed continual learning: Integrate long-tailed learning with continual learning to incrementally learn new knowledge from class-imbalanced data
Laparoscopic Image Analysis for Minimally-Invasive Surgery
Semi/Weakly/Self-Supervised Semantic Segmentation: Learn to segment different organs in laparoscopic surgery image/video with limited labeled data
Human-in-the-loop Learning: Incorporate surgeon's expertise in real-time to refine the model's performance during the live surgeries
|
|