Suppressing the Shortcut Hidden Behind the Gains of VideoLLM Fine-Tuning

Media reports Hugging Face Papers Source language: Chinese

VETO projects supervised fine‑tuning updates onto the orthogonal complement of the text‑only activation subspace, suppressing shortcut learning in VideoLLMs. This improves out‑of‑distribution video reasoning performance.

Full text

VETO suppresses shortcut learning in VideoLLMs by projecting supervised fine-tuning updates onto the orthogonal complement of the text-only activation subspace, improving out-of-distribution video reasoning.

Sources

← Back to Daily AI Report