AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

VAR RL Done Right: Tackling Asynchronous Policy Conflicts in Visual Autoregressive Generation

Shikun Sun, Liao Qu, Huichao Zhang, Yiheng Liu, Yangyang Song, Xian Li, Yi Jiang, Xu Wang, Jia Jia, Daniel K. Du, Xinglong Wu

CVPR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 3 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
apply RL with VMR prefix/suffix split at m=256 vs vanilla GRPO (m=1024) (ablation)NextFlow GRPO · CVTG-2KWord Accuracy: 0.5741 -> 0.6565Improved
move VMR step earlier from m=512 to m=128 (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6351 -> 0.6677Improved
choose m=256 instead of m=128 (default, for compute and mask compatibility) (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6677 -> 0.6565No clear effect
add per-action normalization weighting with exponent 0.6 vs 1.2 (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6602 -> 0.7136Improved
add Mask Propagation (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6855 -> 0.7071Improved
add Mask Propagation (ablation)NextFlow-RL · CVTG-2KCLIPScore: 0.8188 -> 0.8184No clear effect
use K=2 rollouts for VMR instead of K=1 (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6575 -> 0.6821Improved
use K=4 rollouts instead of K=2 (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6821 -> 0.6720Hurt
fine-grained (3:1) alternation of prefix and suffix updates vs coarse-grained (ablation)NextFlow-RL · CVTG-2KWord Accuracy: 0.6778 -> 0.6855Improved
apply OCR-reward RL with VMR/PANW/MPNextFlow · CVTG-2KWord Accuracy: 0.5536 -> 0.7841Improved
apply HPSv3-reward RLNextFlow · HPSv3 benchmark (All)HPS score: 8.43 -> 10.64Improved

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