Idan Daniel Grosbard

The paper uses deep learning models to show that the superior temporal sulcus represents naturalistic social interactions through multimodal integration of visual and auditory information.

This study shows that increasing face image diversity during training improves deep networks' view invariance, sensitivity to critical facial features, and classification performance.

This paper compares self-supervised and supervised models in predicting neural responses to images, finding self-supervision better for fMRI and MEG, with language supervision explaining later stages and image-only supervision capturing early and long latencies.

This study uses deep learning models to show that domain-specific computations are better for fine-grained classification than domain-general ones.

This study shows that combining visual and semantic information improves predictions of human face recognition in memory, suggesting a new visual-semantic representation in human perception.