The paper introduces Rita, a reinforcement learning framework that addresses thinking drift in vision‑language models by enforcing consistency between reasoning and answers. Rita employs two reasoning‑label‑free rewards—thinking and consistency rewards—derived from the conditional probability of reference answers, and uses a difficulty‑aware data filtering strategy to select informative samples for training. Experiments on EgoIntention and RefEgo‑Int benchmarks demonstrate that Rita outperforms both supervised fine‑tuning and vanilla RL‑fine‑tuned approaches.
This paper studies \textbf{thinking--answer consistency} in vision-language models. We focus on Visual Intention Grounding, where a model infers a target object based on a human intention query and pr...
arXiv:2610.02015v1 Announce Type: cross Abstract: Recent advances in LLM reasoning models---driven primarily by the paradigm of post-training via reinforcement learning with verifiable reward (RLVR)-...
arXiv:2606. 29984v1 Announce Type: new Abstract: Reinforcement Learning (RL) is an important paradigm for improving the reasoning capabilities of Vision-Language Models (VLMs).
The paper introduces Stepwise Marginal Information Gain (MIG), an intrinsic process reward that evaluates how each reasoning step of a large language model (LLM) or vision-language model (VLM) improves the likelihood of the reference answer. MIG rewards only new likelihood maxima, preventing duplicate credit, and is combined with outcome, format, and self‑distillation objectives to guide training. Experiments on eight benchmarks show that this method outperforms outcome‑only reinforcement learning and improves accuracy by up to 4.8 points over binary‑reward training, including a 12.6‑point gain on MathVerse and a 12.9‑point advantage on vision‑language transfer at 7B parameters.
arXiv:2606. 11209v1 Announce Type: cross Abstract: Visual question answering increasingly requires multi-step reasoning.
arXiv:2603. 16728v2 Announce Type: replace Abstract: Vision-language models (VLMs) are increasingly deployed in high-stakes settings where reliable uncertainty quantification (UQ) is as important as predictive accuracy.
arXiv:2606. 16122v1 Announce Type: new Abstract: Visual thinking should not only sound right; it should show its evidence.
arXiv:2608. 19669v1 Announce Type: cross Abstract: Latent reasoning has advanced multimodal reasoning through a two-stage training paradigm: (1) a helper image is encoded into latent tokens to teach visual chain-of-thought during a supervised fine-tuning (SFT) stage, and (2) these latent tokens are further refined with reward feedback during a reinforcement learning (RL) stage.
arXiv:2604. 04917v3 Announce Type: replace-cross Abstract: What does it take to build a visual reasoner that works across charts, science, spatial understanding, and open-ended tasks?
SAVOR is a training framework for multimodal large language models that adds token and answer confidence to the output schema, optimises a Group Relative Policy Optimisation objective to penalise calibration error and poor abstention, and uses the learned confidence at inference to revisit visual evidence only when uncertain. Experiments on POPE, HallusionBench, AMBER, and MMHal-Bench with InternVL3-8B and Qwen3-VL-8B backbones show that SAVOR reduces hallucination while maintaining general capability on MME and MMBench, achieving lower Expected Calibration Error than DPO and decoding baselines.
The paper introduces V‑Rubrics, a reinforcement‑learning framework that evaluates vision‑language model responses by breaking them into atomic propositions and scoring them on Visual Faithfulness, Reasoning Consistency, and Instruction Following. Using a fine‑tuned Qwen3‑VL‑8B‑Instruct model and a newly created 50K‑example V‑Rubrics dataset, the authors demonstrate that rubric‑based GRPO outperforms both a shared SFT baseline and an answer‑only GRPO, especially on knowledge‑oriented and visually grounded reasoning tasks.
arXiv:2608. 03875v1 Announce Type: cross Abstract: Designing effective reward functions remains a major bottleneck in Reinforcement Learning (RL).