Back to AIBriefs
AnalysisAI Models

ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation

ViCuR uses visual cues as a recoverable privilege signal to improve multimodal on-policy distillation. The method trains a student on its own policy trajectories under teacher supervision, enhancing reasoning performance.

·
6 days ago
ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation — AIBriefs