Research map / Vision-language and multimodal / Multimodal LLMs
DEX-AR: A Dynamic Explainability Method for Autoregressive Vision-Language Models
Walid Bousselham, Angie Boggust, Hendrik Strobelt, Hilde Kuehne
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 |
|---|---|---|---|
| add dynamic head filtering (max-based) (ablation) | DEX-AR · PascalVOC (LLaVA) | SNR: 1.64 -> 3.64 | Improved |
| add dynamic head filtering (max-based) (ablation) | DEX-AR · PascalVOC (BakLLaVA) | SNR: 5.27 -> 6.02 | Improved |
| score heads by max instead of average gradient (ablation) | DEX-AR · PascalVOC (LLaVA) | SNR: 1.09 -> 3.64 | Improved |
| score heads by top-0.05 gradients instead of max (ablation) | DEX-AR · PascalVOC (LLaVA) | SNR: 3.64 -> 3.35 | Hurt |
| add head filtering alone (ablation) | DEX-AR · PascalVOC-QA (LLaVA-1.5-7B) | SNR: 9.16 -> 16.84 | Improved |
| add filler-token filtering alone (ablation) | DEX-AR · PascalVOC-QA (LLaVA-1.5-7B) | SNR: 9.16 -> 89.29 | Improved |
| add head filtering on top of filler filtering (ablation) | DEX-AR · PascalVOC-QA (LLaVA-1.5-7B) | SNR: 89.29 -> 96.12 | Improved |
| add head filtering on top of filler filtering (ablation) | DEX-AR · PascalVOC-QA (LLaVA-1.5-7B) | MSE: 0.11 -> 0.12 | Hurt |
| add ReLU on attention gradients (ablation) | DEX-AR · PascalVOC (LLaVA) | IoU: 31.56 -> 36.34 | Improved |
| use intermediate layers instead of last layer only (ablation) | DEX-AR · PascalVOC (LLaVA) | IoU: 33.05 -> 36.34 | Improved |
| use intermediate layers instead of last layer only (ablation) | DEX-AR · PascalVOC (Florence-2) | IoU: 26.91 -> 32.48 | Improved |
| add ReLU on attention gradients (ablation) | DEX-AR · PascalVOC (PaliGemma) | IoU: 22.90 -> 23.55 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GradCAM — Grad-CAM: Visual Explanations From Deep Networks via Gradient-Based Localization (ICCV 2017) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Data-efficient Large Vision Models through Sequential Autoregression (ICML 2024 · similarity 0.90)
- Where MLLMs Attend and What They Rely On: Explaining Autoregressive Token Generation (CVPR 2026 · similarity 0.90)
- Language Model as Visual Explainer (NeurIPS 2024 · similarity 0.89)
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.88)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.88)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.88)
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.
