Tag
The paper examines how vision-language models handle conflicts between in-context and parametric knowledge, revealing asymmetric biases where models favor in-context information for text entities but parametric information for image entities due to slower visual processing.
This paper investigates the use of conversational temporal dynamics (turn-pair timing) as a lightweight modality for automatic depression detection from dyadic clinical interviews, showing that a compact 24-dimensional timing module achieves strong performance and complements standard acoustic and semantic features when fused.