BWhat changed
Added local text-to-text attention within dedicated visual fusion layers, retaining cross-attention-like scaling while narrowing the tested performance gap to full visual-token insertion.
Scientific scope
Added local text-to-text attention within dedicated visual fusion layers, retaining cross-attention-like scaling while narrowing the tested performance gap to full visual-token insertion. Inherits cross-attention and blockwise attention. Efficiency and accuracy comparisons are configuration-dependent; continuous visual updates do not give unbounded visual memory. Original title uses via Self-Attention, unlike the later index title.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits cross-attention and blockwise attention. Efficiency and accuracy comparisons are configuration-dependent; continuous visual updates do not give unbounded visual memory. Original title uses via Self-Attention, unlike the later index title.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- CASA: Cross-Attention via Self-Attention for Efficient Vision-Language Fusion
- Kyutai scientific publications index
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction