AI Paper Atlas

Research map / Video / Video-language understanding

Causality Matters: How Temporal Information Emerges in Video Language Models

Yumeng Shi, Quanyu Long, Yin Wu, Wenya Wang

AAAI 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 4 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
remove positional encodings from a single layer after layer 1 (ablation)Qwen2.5-VL-7B · TempCompassGT token prob: within +-2%No clear effect
remove positional encodings in the first layer (ablation)Qwen2.5-VL-7B · TempCompassGT token prob: up to -60%Hurt
shuffle video token position IDs in layer 1 (ablation)Qwen2.5-VL-7B · TempCompass MCQGT token prob: +0.07%No clear effect
shuffle query token position IDs in layer 1 (ablation)Qwen2.5-VL-7B · TempCompass MCQGT token prob: -53.09%Hurt
reverse temporal position IDs only (ablation)Qwen2.5-VL-7B · TempCompassGT token prob: -1.13% to -1.76%No clear effect
reverse frame order keeping original position IDs (ablation)Qwen2.5-VL-7B · TempCompass MCQGT token prob: -15.45%Hurt
block final token attention to query tokens (ablation)Qwen2.5-VL-7B · TempCompass MCQGT token prob: exceeds -40%Hurt
block final token attention to frame tokens (ablation)Qwen2.5-VL-7B · TempCompassGT token probNo clear effect
block inter-frame attention (ablation)Qwen2.5-VL-7B · TempCompassGT token prob: -4% to -6% near layer 10Hurt
restrict query attention to Frame 4 only instead of Frame 1 (ablation)Qwen2.5-VL-7B · TempCompass MCQGT token prob: -8% (Frame 1) vs below 2% (Frame 4)Improved
disable frame-to-frame attention in layers 20-28Qwen2.5-VL-7B · NExT-QAaccuracy: 75.1 -> 75.2No clear effect
remove frame tokens from KV cache in layers 20-28Qwen2.5-VL-7B · TempCompass Yes/Noaccuracy: 70.8 -> 70.7No clear effect

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