AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Probing Cross-modal Information Hubs in Audio-Visual LLMs

Jihoo Jung, Chaeyoung Jung, Ji-Hoon Kim, Joon Son Chung

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
patch sink tokens (N=2) instead of random tokens (ablation)Qwen2.5-Omni (7B) · Audio-dominant causal tracingIE_clean: 4.24 -> 6.24Improved
patch sink tokens (N=2) instead of random tokens (ablation)Qwen2.5-Omni (7B) · Video-dominant causal tracingIE_clean: 4.56 -> 5.47Improved
patch sink tokens (N=2) instead of object tokens (ablation)Qwen2.5-Omni (7B) · Audio-dominant causal tracingIE_clean: 5.04 -> 6.24Improved
patch cross-modal sink tokens instead of unimodal sink tokens (ablation)Qwen2.5-Omni (7B) · Audio-dominant causal tracingIE_clean: 0.65 -> 5.58Improved
patch cross-modal sink tokens instead of unimodal sink tokens (ablation)video-SALMONN-o1 (7B) · Audio-dominant causal tracingIE_clean: 7.25 -> 21.30Improved
patch cross-modal sink tokens instead of unimodal sink tokens (ablation)Qwen2.5-Omni (7B) · Video-dominant causal tracingIE_clean: 1.93 -> 3.03Improved
patch cross-modal sink tokens instead of all sink tokens (ablation)Qwen2.5-Omni (7B) · Audio-dominant causal tracingIE_clean: 6.24 -> 5.58No clear effect
apply Adaptive Sink-Guided DecodingQwen2.5-Omni (7B) · VGGSound-AnimalCHAIR Cs: 48.21 -> 36.91Improved
apply Adaptive Sink-Guided Decodingvideo-SALMONN-o1 (7B) · VGGSound-AnimalCHAIR Cs: 37.74 -> 25.07Improved
apply Adaptive Sink-Guided DecodingQwen2.5-Omni (7B) · AudioSetCHAIR Cs: 8.92 -> 8.54Improved
apply Adaptive Sink-Guided DecodingQwen2.5-Omni (7B) · VGGSound-Animalcaption F1: 55.24 -> 52.44Hurt
apply PAIQwen2.5-Omni (7B) · VGGSound-AnimalCHAIR Cs: 48.21 -> 51.24Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy