English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Purpose: In curriculum learning, the idea is to train on easier samples firstand gradually increase the difficulty, while in self-paced learning, a pacingfunction defines the speed to adapt the training progress. While both methodsheavily rely on the ability to score the difficulty of data samples, an optimalscoring function is still under exploration. Methodology: Distillation is aknowledge transfer approach where a teacher network guides a student network byfeeding a sequence of random samples. We argue that guiding student networkswith an efficient curriculum strategy can improve model generalization androbustness. For this purpose, we design an uncertainty-based paced curriculumlearning in self distillation for medical image segmentation. We fuse theprediction uncertainty and annotation boundary uncertainty to develop a novelpaced-curriculum distillation (PCD). We utilize the teacher model to obtainprediction uncertainty and spatially varying label smoothing with Gaussiankernel to generate segmentation boundary uncertainty from the annotation. Wealso investigate the robustness of our method by applying various types andseverity of image perturbation and corruption. Results: The proposed techniqueis validated on two medical datasets of breast ultrasound image segmentationand robotassisted surgical scene segmentation and achieved significantly betterperformance in terms of segmentation and robustness. Conclusion: P-CD improvesthe performance and obtains better generalization and robustness over thedataset shift. While curriculum learning requires extensive tuning ofhyper-parameters for pacing function, the level of performance improvementsuppresses this limitation.

Paced-Curriculum Distillation with Prediction and Label Uncertainty for Image Segmentation