Andrew P. King

Marginal loss outperforms other loss functions in complex echocardiography segmentation tasks with multiple missing labels, revealing a new frontier in handling partially labelled data.

Meta-learning outperformed other strategies in cardiac motion estimation, achieving superior adaptation trajectories over time.

Even a standard U-Net can maintain surprisingly strong performance in echocardiography segmentation despite significant ground truth errors, but a Variance of Gradients approach can boost performance further by detecting and correcting those errors during training.

Semantic segmentation models can get object boundaries right but still confidently mislabel the object's identity when category and scene are spuriously correlated.