Mechanistic Insights into Functional Sparsity in Multimodal LLMs via CoRe Heads

• URL: http://arxiv.org/abs/2606.05843v1
• Date: Thu, 04 Jun 2026 08:18:31 GMT
• Title: Mechanistic Insights into Functional Sparsity in Multimodal LLMs via CoRe Heads
• Abstract summary: We present an in-depth interpretability study that uncovers a profound structural property within Multimodal Large Language Models (MLLMs)<n>We identify and characterize a highly specialized subset of attention heads, referred to as Context-aware Retrieval (CoRe) heads.<n>Our findings reveal a structural principle of functional sparsity within MLLMs, refining the current understanding of mechanistic interpretability.
• Score: 48.341678148905466
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: While Multimodal Large Language Models (MLLMs) demonstrate remarkable proficiency on complex vision-language tasks, the mechanisms by which they extract query-relevant visual features from complex, noisy contexts remain opaque. In this paper, we present an in-depth interpretability study that uncovers a profound structural property within MLLMs: functional sparsity in cross-modal retrieval. Leveraging a token-level metric termed Retrieval Attention Mass (RAM), we identify and characterize a highly specialized subset of attention heads, referred to as Context-aware Retrieval (CoRe) heads. Across diverse visual domains and model scales, we observe a clear functional division: CoRe heads act as dedicated information extractors, while most other heads distribute attention over broader contextual regions. Causal interventions further demonstrate the necessity of these specialized heads. Ablating only the top 5% of CoRe heads causes significant degradation in multimodal reasoning performance, whereas ablating lower-ranked heads has minimal effect. Moreover, acceleration experiments validate the utility of CoRe heads, showing that leveraging this localized sparsity significantly accelerates inference while maintaining robust task performance. Our findings reveal a structural principle of functional sparsity within MLLMs, refining the current understanding of mechanistic interpretability and laying a theoretical foundation that can inspire future architecture design and model optimization.

Related papers

• Large Vision-Language Models Get Lost in Attention [51.851592109135716]
  We propose a unified framework grounded in information theory and geometry to quantify the geometric and entropic nature of residual updates.<n>Applying this unified framework reveals a fundamental functional decoupling: Attention acts as a subspace-language operator focused on reconfiguration, whereas FFNs serve as subspace-language operators driving semantic innovation.
  arXiv  Detail & Related papers  (2026-05-07T04:45:52Z)
• Predictive Regularization Against Visual Representation Degradation in Multimodal Large Language Models [84.94288033791346]
  We conduct a detailed diagnostic analysis to unveil a pervasive issue: visual representation degradation in MLLMs.<n>We attribute this phenomenon to a visual sacrifice driven by the singular text-generation objective, where the model compromises its visual fidelity to optimize for answer generation.<n>We propose Predictive Regularization to force degraded intermediate features to predict initial visual features, thereby maintaining the inherent visual attributes of the MLLM's internal representations.
  arXiv  Detail & Related papers  (2026-03-21T13:10:37Z)
• Attention in Space: Functional Roles of VLM Heads for Spatial Reasoning [43.03674069044073]
  We introduce CogVSR, a dataset that decomposes complex spatial reasoning questions into step-by-step subquestions.<n>We develop a probing framework to identify and characterize attention heads specialized for these functions.<n>We propose methods to activate latent spatial heads, improving spatial understanding.
  arXiv  Detail & Related papers  (2026-03-21T05:36:12Z)
• Investigating The Functional Roles of Attention Heads in Vision Language Models: Evidence for Reasoning Modules [76.21320451720764]
  We introduce CogVision, a dataset that decomposes complex multimodal questions into step-by-step subquestions.<n>Using a probing-based methodology, we identify attention heads that specialize in these functions and characterize them as functional heads.<n>Our analysis reveals that these functional heads are universally sparse, vary in number and distribution across functions, and mediate interactions and hierarchical organization.
  arXiv  Detail & Related papers  (2025-12-11T05:42:53Z)
• Cognitive Mirrors: Exploring the Diverse Functional Roles of Attention Heads in LLM Reasoning [54.12174882424842]
  Large language models (LLMs) have achieved state-of-the-art performance in a variety of tasks, but remain largely opaque in terms of their internal mechanisms.<n>We propose a novel interpretability framework to systematically analyze the roles and behaviors of attention heads.
  arXiv  Detail & Related papers  (2025-12-03T10:24:34Z)
• From Compression to Expression: A Layerwise Analysis of In-Context Learning [24.45948310980883]
  In-context learning (ICL) enables large language models to adapt to new tasks without weight updates by learning from demonstration sequences.<n>We conduct a statistical geometric analysis of ICL representations to investigate how task-specific information is captured across layers.<n>Our findings reveal an intriguing layerwise dynamic in ICL, highlight how structured representations emerge within LLMs, and showcase that analyzing internal representations can facilitate a deeper understanding of model behavior.
  arXiv  Detail & Related papers  (2025-05-22T22:22:03Z)
• Cantor: Inspiring Multimodal Chain-of-Thought of MLLM [83.6663322930814]
  We argue that converging visual context acquisition and logical reasoning is pivotal for tackling visual reasoning tasks. We propose an innovative multimodal CoT framework, termed Cantor, characterized by a perception-decision architecture. Our experiments demonstrate the efficacy of the proposed framework, showing significant improvements in multimodal CoT performance.
  arXiv  Detail & Related papers  (2024-04-24T17:59:48Z)

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.