The Flow has not summarised this story yet — read it at arXiv Computation and Language.
The paper introduces RECAP, a training method that enhances vision‑language models by incorporating temporally localized reactivation of image evidence—called visual revisit—into the reinforcement learning process. Visual revisit peaks are shown to predict correctness and causally improve performance; masking them hurts accuracy more than masking non‑peaks, while adding revisit peaks to failed reasoning partially restores correct answers. RECAP applies revisit‑conditioned advantage estimation, dynamic discounting, and gated value prediction to propagate final rewards through visually important reasoning steps, achieving significant gains on HallusionBench and other benchmarks without extra rewards or inference‑time changes.
VGIQA is a new image quality assessment model that provides both local and global quality evaluations. It uses a two‑stage training process—starting with a cold‑start phase and then reinforcement learning guided by local quality reward functions. The model demonstrates superior performance on the newly introduced VGSR‑2K dataset, outperforming existing open‑source multi‑modal large language models such as Qwen2.5‑VL‑72B‑Instruct and Qwen3‑VL‑32B‑Instruct, and it also improves degradation perception tasks.
arXiv:2609.39168v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet existing...
arXiv:2510.23482v2 Announce Type: replace Abstract: Recent large vision-language models (LVLMs) can generate vision-text multimodal chain-of-thought (MCoT) traces after reinforcement fine-tuning (RFT...
The paper introduces Perception and Reasoning Attribution (PARA), a method that quantifies how much of a vision‑language model’s errors can be fixed by better perception versus stronger reasoning. PARA reveals that reinforcement learning (RL) benefits more from improving reasoning than perception, and motivates a new group‑relative attention distillation objective called GRAD. GRAD directly guides weaker rollouts toward the visual attention patterns of the best rollouts, consistently boosting performance across multiple model scales and benchmarks.
VGIQA is a new image quality assessment model that provides both local and global quality evaluations. It uses a two-stage training process—initial cold-start followed by reinforcement learning—with local quality reward functions. The model outperforms existing large multimodal language models on the newly introduced VGSR-2K dataset and improves performance on degradation perception tasks.