AI Paper Atlas

Research map / Large language models / Alignment and preferences

Linguistic Relative Policy Optimization for Video Anomaly Reasoning

Jiaxu Leng, Jiankang Zheng, Mengjingcheng Mo, Zhanjie Wu, Haosheng Chen, Ji Gan, Xinbo Gao

ICML 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 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
inject LRPO-learned linguistic experience into context (ablation)InternVL3 5-8B · XD-ViolenceAP (%): 59.93 -> 73.17Improved
learn experience from group of trajectories instead of single trajectory (ablation)LRPO · XD-ViolenceAP (%): 66.78 -> 73.17Improved
add scenario experience on top of general experience (accuracy reward) (ablation)LRPO · XD-ViolenceAP (%): 68.48 -> 70.58Improved
add anomaly preference reward (ablation)LRPO · XD-ViolenceAP (%): 68.48 -> 69.78Improved
add anomaly temporal dependency reward (ablation)LRPO · XD-ViolenceAP (%): 68.48 -> 70.06Improved
combine all three rewards (ablation)LRPO · XD-ViolenceAP (%): 68.48 -> 71.91Improved
random scenario experience injection (ablation)LRPO · XD-ViolenceAP (%): 71.91 -> 71.61Hurt
visual-semantic retrieval instead of visual-only retrieval (ablation)LRPO scenario injection · XD-ViolenceAP (%): 72.54 -> 73.17Improved
progressively refine experience during optimization (ablation)LRPO · XD-ViolenceAP (%): 67.63 -> 73.17Improved
increase scenario experience ratio at inference (ablation)LRPO · XD-ViolenceAP (%): 73.17 -> 74.52Improved
scale scenario experience during experience learning (ablation)LRPO · XD-ViolenceAP (%): 73.17 -> 71.20Hurt
use full training set instead of 100 videos (ablation)LRPO · XD-ViolenceAP (%): 73.17 -> 74.09No 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