Kerstin Ritter

Acoustic noise, even when not diagnostically informative, can systematically manipulate Alzheimer's predictions in SSL models, revealing a critical vulnerability in current speech-based assessments.

Omitted variable bias in deep learning can be effectively controlled, leading to unbiased predictions and improved model performance even in the presence of confounding variables.

ICON decomposition reveals the true reliance of deep models on concepts, debunking misleading correlations that traditional methods often overlook.

Layer selection for speech-based PD detection is more about the dataset than the model architecture, revealing a critical flaw in current approaches.

SSL embeddings, typically superior for speech tasks, surprisingly lose their edge to hand-crafted acoustic features when classifying mild cognitive impairment from speech, challenging assumptions about representation learning in clinical applications.