Modality-Informed Reciprocal Reasoning Optimization (MIRROR), a reinforcement learning approach for improving multimodal reasoning via self supervision, is developed and improves over standard RL and yields more accurate and consistent behavior across modalities.
Abstract
Unlike large language models (LLMs) that exhibit strong reasoning capabilities, vision-language models (VLMs) struggle with visual reasoning, even on geometry problems that admit equivalent text, diagram, and combined diagram+text views. We show that these views often elicit different behaviors: a model may solve a problem from text but fail on the corresponding diagram, or succeed visually while failing textually. This inconsistency suggests that different views expose complementary reasoning paths and failure modes that standard multimodal post-training does not fully exploit. To study and exploit this phenomenon, we construct ODA-Data, a high-quality paired multimodal geometry dataset with text-dominant, image-dominant, and combined image+text views of the same problems, together with splits for training and evaluating modality-dependent reasoning behaviors. We then develop Modality-Informed Reciprocal Reasoning Optimization (MIRROR), a reinforcement learning approach for improving multimodal reasoning via self supervision. For each problem, MIRROR evaluates the model under all views, selects the best-performing view as a teacher, and trains other views with a reverse-KL objective towards the teacher. Across reasoning benchmarks that evaluate on geometry problems, MIRROR improves over standard RL and yields more accurate and consistent behavior across modalities
Multimodal queries can require different types of reasoning. Some can be answered via perceptual reasoning, extracting information directly from the visual signal, while others require compositional reasoning that combines observations or deliberative reasoning that evaluates competing hypotheses. However, many existing methods apply a uniform reasoning strategy across queries, leading to unnecessary computation on simple tasks and insufficient reasoning on complex ones. We introduce Video-FLAIR, a training framework that learns to select the appropriate reasoning mode for each query using reinforcement learning. During training, the model generates responses under all three modes for the same prompt, enabling direct comparison. A composite reward compares these responses to favor the most effective one based on correctness, grounding, and cost, while discouraging unsupported or misaligned deliberation. This yields a supervision signal for learning adaptive reasoning without per-query annotations. Video-FLAIR improves accuracy over the Qwen2.5-VL base model by +5.4 on MathVista, +4.8 on Video-Holmes, and +4.8 on Video-MMMU, while reducing average token usage to 95 compared to 417 for always-thinking baselines.
TRAM (TRajectory-derived Auxiliary Memory), a training-free method that augments standard decoding with an auxiliary memory pathway derived from the model's own reasoning trajectory, shows that TRAM improves performance over vanilla decoding on mathematical, scientific, and general visual reasoning tasks without additional training.
Kang Liu, Zijing Wang, Yongkang Liu et al.· 0 citations
For the coarse attributes the authors study, MLLMs encode the visual evidence but cannot reliably control their reliance on it, indicating that for the coarse attributes they study, MLLMs cannot reliably control their reliance on it.
Jiaang Li, Chengzu Li, Zhaochong An et al.· 0 citations
Cross-Modal Alignment Regularization (CMAR) is proposed, a method designed to improve LLMs by aligning their internal representations with those of vision models during training by bringing the internal representations of the language and vision models closer together.
Yulu Gan, Kaiya Ivy Zhao, Tomaso A. Poggio et al.· 0 citations
Vision-Language Models (VLMs) have achieved remarkable progress in aligning visual and textual information; however, their inference processes remain largely implicit, end-to-end, and weakly structured. As a result, even state-of-the-art models often struggle with logical consistency, spatial reasoning, multi-entity binding, and robustness to occlusion or viewpoint changes, limiting their reliability in scientific, industrial, and safety-critical applications. To address these limitations, we propose Structure-Aware Visual Reasoning (SAVR), a conceptual framework that augments VLMs with explicit representation and constraint-based reasoning components. SAVR decomposes visual inference into three interpretable stages: (i) an entity–attribute–relation (EAR) representation, (ii) constraint-aware reasoning over structured representations, and (iii) integration of structured predictions with VLM-generated outputs. This formulation treats visual reasoning as an explicit process that enforces spatial, logical, and physical constraints over grounded entities, thereby improving consistency, transparency, and controllability. Furthermore, SAVR unifies several previously fragmented research directions-including scene graphs, object-centric learning, neuro-symbolic reasoning, and spatially grounded VLMs-into a coherent architectural perspective. Through this synthesis, we clarify how explicit structural representations can systematically mitigate common failure modes of end-to-end VLMs. While this work is conceptual and does not include empirical evaluation, it provides rigorous problem formulation, design rationale, and comparative analysis that establish a foundation for future structure-aware multimodal reasoning systems.
