Sam Gijsen

Conditional flow models hallucinate non-additive scientific interactions even when achieving high aggregate fidelity, failing to generalize interaction structure when condition patterns diverge.

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.

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

Zero-shot generation of fMRI dynamics for unseen cognitive tasks could revolutionize experimental neuroscience by enabling in-silico testing of novel hypotheses.

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.