Julia E. Vogt

Concept interventions now propagate to residual pathways, enabling effective steering even when training concepts are incomplete.

Linear-complexity state-space modeling eliminates the trade-off between receptive field size and inference speed in physiological signal denoising, yielding performance advantages that actually widen on longer sequences and context-dependent ST/T pathologies.

Multi-view attention in masked autoencoders unlocks surprisingly robust and transferable representations for echocardiography, even across significant anatomical differences between adults and children.