Learning When to Look: A Disentangled Curriculum for Strategic Perception in Multimodal Reasoning

• URL: http://arxiv.org/abs/2512.17227v1
• Date: Fri, 19 Dec 2025 04:25:57 GMT
• Title: Learning When to Look: A Disentangled Curriculum for Strategic Perception in Multimodal Reasoning
• Authors: Siqi Yang, Zilve Gao, Haibo Qiu, Fanfan Liu, Peng Shi, Zhixiong Zeng, Qingmin Liao, Lin Ma,
• Abstract summary: A critical failure mode is "visual forgetting", where models progressively lose visual grounding as reasoning extends.<n>We propose a novel curriculum-based framework to disentangle these skills.<n>Our contributions include the formalization of these two deficiencies and the development of a principled, two-stage framework to address them.
• Score: 35.206515263989566
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal Large Language Models (MLLMs) demonstrate significant potential but remain brittle in complex, long-chain visual reasoning tasks. A critical failure mode is "visual forgetting", where models progressively lose visual grounding as reasoning extends, a phenomenon aptly described as "think longer, see less". We posit this failure stems from current training paradigms prematurely entangling two distinct cognitive skills: (1) abstract logical reasoning "how-to-think") and (2) strategic visual perception ("when-to-look"). This creates a foundational cold-start deficiency -- weakening abstract reasoning -- and a strategic perception deficit, as models lack a policy for when to perceive. In this paper, we propose a novel curriculum-based framework to disentangle these skills. First, we introduce a disentangled Supervised Fine-Tuning (SFT) curriculum that builds a robust abstract reasoning backbone on text-only data before anchoring it to vision with a novel Perception-Grounded Chain-of-Thought (PG-CoT) paradigm. Second, we resolve the strategic perception deficit by formulating timing as a reinforcement learning problem. We design a Pivotal Perception Reward that teaches the model when to look by coupling perceptual actions to linguistic markers of cognitive uncertainty (e.g., "wait", "verify"), thereby learning an autonomous grounding policy. Our contributions include the formalization of these two deficiencies and the development of a principled, two-stage framework to address them, transforming the model from a heuristic-driven observer to a strategic, grounded reasoner. \textbf{Code}: \url{https://github.com/gaozilve-max/learning-when-to-look}.

Related papers

• MentisOculi: Revealing the Limits of Reasoning with Mental Imagery [63.285794947638614]
  We develop MentisOculi, a suite of multi-step reasoning problems amenable to visual solution.<n> evaluating visual strategies ranging from latent tokens to explicit generated imagery, we find they generally fail to improve performance.<n>Our findings suggest that despite their inherent appeal, visual thoughts do not yet benefit model reasoning.
  arXiv  Detail & Related papers  (2026-02-02T18:49:06Z)
• Analyzing Reasoning Consistency in Large Multimodal Models under Cross-Modal Conflicts [74.47786985522762]
  We identify a critical failure mode termed textual inertia, where models tend to blindly adhere to the erroneous text while neglecting conflicting visual evidence.<n>We propose the LogicGraph Perturbation Protocol that structurally injects perturbations into the reasoning chains of diverse LMMs.<n>Results reveal that models successfully self-correct in less than 10% of cases and predominantly succumb to blind textual error propagation.
  arXiv  Detail & Related papers  (2026-01-07T16:39:34Z)
• Guiding the Inner Eye: A Framework for Hierarchical and Flexible Visual Grounded Reasoning [6.800544911407401]
  GRiP (Guided Reasoning and Perception) is a novel training framework for visual grounded reasoning.<n> GRiP cultivates robust and flexible visual grounded reasoning by explicitly guiding the model's perceptual focus and logical pathways.<n> GRiP achieves state-of-the-art results among open-source models on the highly challenging TreeBench and V* Bench.
  arXiv  Detail & Related papers  (2025-11-27T07:18:25Z)
• Think Visually, Reason Textually: Vision-Language Synergy in ARC [94.15522924153264]
  ARC-AGI is a rigorous testbed for conceptual rule induction and transfer to novel tasks.<n>Most existing methods treat ARC-AGI as a purely textual reasoning task, overlooking the fact that humans rely heavily on visual abstraction.<n>We introduce two synergistic strategies: Vision-Language Synergy Reasoning (VLSR) and Modality-Switch Self-Correction (MSSC)<n>Our findings suggest that unifying visual abstraction with linguistic reasoning is a crucial step toward achieving generalizable, human-like intelligence.
  arXiv  Detail & Related papers  (2025-11-19T18:59:04Z)
• BLINK-Twice: You see, but do you observe? A Reasoning Benchmark on Visual Perception [67.89135437537179]
  We introduce BLINK-Twice, a vision-centric reasoning benchmark grounded in challenging perceptual tasks.<n>Instead of relying on external knowledge, our tasks require models to reason from visual content alone.<n>Compared to prior perception benchmarks, it moves beyond shallow perception and requires fine-grained observation and analytical reasoning.
  arXiv  Detail & Related papers  (2025-10-10T13:14:13Z)
• More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models [17.431298099935344]
  Reasoning has emerged as a pivotal capability in Large Language Models (LLMs)<n>Recent research has sought to extend reasoning to Vision-Language Models (VLMs)<n>Our study uncovers the dual nature of multimodal reasoning, leading to recognition failures on otherwise basic visual questions.<n>We propose Vision-Anchored Policy Optimization (VAPO), a simple yet effective method that explicitly steers the reasoning process toward visually grounded trajectories.
  arXiv  Detail & Related papers  (2025-09-30T06:37:47Z)
• Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models [67.87579664988199]
  TON is a two-stage training strategy for vision-language models (VLMs)<n>It introduces a think-or-not format that serves as a cold start for selective reasoning.<n>TON can reduce the completion length by up to 90% compared to vanilla GRPO.
  arXiv  Detail & Related papers  (2025-05-22T16:13:29Z)
• Embodied-R: Collaborative Framework for Activating Embodied Spatial Reasoning in Foundation Models via Reinforcement Learning [58.86928947970342]
  Embodied-R is a framework combining large-scale Vision-Language Models for perception and small-scale Language Models for reasoning.<n>After training on only 5k embodied video samples, Embodied-R with a 3B LM matches state-of-the-art multimodal reasoning models.<n>Embodied-R also exhibits emergent thinking patterns such as systematic analysis and contextual integration.
  arXiv  Detail & Related papers  (2025-04-17T06:16:11Z)
• A Cognitive Paradigm Approach to Probe the Perception-Reasoning Interface in VLMs [3.2228025627337864]
  This paper introduces a structured evaluation framework to dissect the perception-reasoning interface in Vision-Language Models (VLMs)<n>We propose three distinct evaluation paradigms, mirroring human problem-solving strategies.<n>Applying this framework, we demonstrate that CA, leveraging powerful language models for reasoning over rich, independently generated descriptions, achieves new state-of-the-art (SOTA) performance.
  arXiv  Detail & Related papers  (2025-01-23T12:42:42Z)
• Conceptual and Unbiased Reasoning in Language Models [98.90677711523645]
  We propose a novel conceptualization framework that forces models to perform conceptual reasoning on abstract questions. We show that existing large language models fall short on conceptual reasoning, dropping 9% to 28% on various benchmarks. We then discuss how models can improve since high-level abstract reasoning is key to unbiased and generalizable decision-making.
  arXiv  Detail & Related papers  (2024-03-30T00:53:53Z)
• Does Deep Learning Learn to Abstract? A Systematic Probing Framework [69.2366890742283]
  Abstraction is a desirable capability for deep learning models, which means to induce abstract concepts from concrete instances and flexibly apply them beyond the learning context. We introduce a systematic probing framework to explore the abstraction capability of deep learning models from a transferability perspective.
  arXiv  Detail & Related papers  (2023-02-23T12:50:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.