Research map / Vision-language and multimodal / Cross-modal learning
Structural Graph Probing of Vision-Language Models
Haoyu He, Yue Zhuo, Yu Zheng, Qi R. Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use GCN probe on correlation graph instead of linear probe | graph probing · CLEVR | probe accuracy: 0.920 -> 0.963 (Qwen2.5-VL-3B) | Improved |
| use GCN probe instead of linear probe | graph probing · MMMU, MMMU-Pro, BLINK, EMMA | probe accuracy: MMMU LLaVA 0.314 -> 0.279; BLINK LLaVA 0.647 -> 0.592 | Mixed |
| use GCN probe instead of linear probe for counting regression | graph probing · CLEVR counting | MSE: 0.605 -> 0.379 (LLaVA-1.5-7B) | Improved |
| sweep graph sparsity from 0.01 to 0.20 (ablation) | graph probe · probing accuracy | accuracy | No clear effect |
| use graph probe instead of text-only baselines | hallucination detection · MHaluBench | accuracy: 0.654 -> 0.910 (Qwen2.5-VL-3B, vs Emb. Avg.) | Improved |
| select hubs by graph degree instead of activation magnitude or random for zeroing (ablation) | neuron ablation · TDIUC and CLEVR | accuracy drop | Improved |
| OPPOSITE edge intervention vs IDENTICAL on top neuron pair (ablation) | edge-level intervention · color and counting | accuracy: Qwen2.5-VL-3B drops -54.8 and -63.9 | Hurt |
| scale hub neurons up or down (ablation) | hub perturbation · color and counting | accuracy | Hurt |
Closest papers by meaning
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.88)
- Neural networks are more modular than single neurons suggest (NeurIPS 2026 · similarity 0.87)
- Neuron-based Multifractal Analysis of Neuron Interaction Dynamics in Large Models (ICLR 2025 · similarity 0.87)
- The Underappreciated Power of Vision Models for Graph Structural Understanding (NeurIPS 2025 · similarity 0.87)
- Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs (NeurIPS 2025 · similarity 0.87)
- Pathways of Visual Information Flow in Vision-Language Models (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
