AI Paper Atlas

Research map / Video / Video-language understanding

ReaSon: Reinforced Causal Search with Information Bottleneck for Video Understanding

Yuan Zhou, Litao Hua, Shilong Jin, Wentao Huang, Haoran Duan

AAAI 2026 · Video › Video-language understanding · approach: Event and anomaly reasoning in video · 2 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
add cycle consistency reward to answer reward (ablation)ReaSon (LLaVA-Video-7B, 8 frames) · NExT-QA valAvg accuracy: 80.1 -> 80.5Improved
add counterfactual reward on top of answer and cycle rewards (ablation)ReaSon (LLaVA-Video-7B, 8 frames) · NExT-QA valAvg accuracy: 80.5 -> 81.4Improved
add cycle consistency reward to answer reward (ablation)ReaSon (LLaVA-Video-7B, 8 frames) · EgoSchema subsetaccuracy: 66.0 -> 68.2Improved
add counterfactual reward on top of answer and cycle rewards (ablation)ReaSon (LLaVA-Video-7B, 8 frames) · EgoSchema subsetaccuracy: 68.2 -> 69.0Improved
add ReaSon frame selection to GPT-4oGPT-4o (8 frames) · NExT-QA valAvg accuracy: 72.0 -> 77.6Improved
add ReaSon frame selection to Qwen2.5-VL-7BQwen2.5-VL-7B (8 frames) · EgoSchema subsetaccuracy: 65.8 -> 68.0Improved
add ReaSon frame selection to LLaVA-Video-7BLLaVA-Video-7B (8 frames) · EgoSchema subsetaccuracy: 65.2 -> 69.0Improved
add ReaSon frame selection to GPT-4oGPT-4o (8 frames) · Video-MME (short videos)accuracy: 55.7 -> 65.9Improved
use ReaSon instead of T* with GPT-4o on long videosGPT-4o (32 frames) · Video-MME long videosaccuracy: 59.3 -> 58.2Hurt
add ReaSon to GPT-4oGPT-4o (32 frames) · Video-MME (overall)Overall accuracy: 61.8 -> 66.4Improved
use ReaSon with LLaVA-Video-7B instead of uniform GPT-4o at 8 framesLLaVA-Video-7B (8 frames) · Video-MME (overall)Overall accuracy: 53.3 vs T* 56.5 (GPT-4o)Hurt

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