Research map / Vision-language and multimodal / Human and embodied modeling
Causal Attention for Vision-Language Tasks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace BUTD attention with CATT | BUTD · MS COCO Karpathy split | CIDEr-D: 125.3 -> 128.3 | Helped |
| Replace Transformer attention with CATT | Transformer · MS COCO Karpathy split | CIDEr-D: 128.5 -> 131.7 | Helped |
| Replace BUTD attention with CATT | BUTD · VQA2.0 test-std | accuracy: 65.67 -> 67.26 | Helped |
| Replace Transformer attention with CATT | Transformer · VQA2.0 test-std | accuracy: 69.82 -> 71.27 | Helped |
| Replace LXMERT† attention with CATT | LXMERT† · VQA2.0 test-std | accuracy: 72.18 -> 73.04 | Helped |
| Replace LXMERT† attention with CATT | LXMERT† · GQA test-std | accuracy: 59.94 -> 61.17 | Helped |
| Replace LXMERT† attention with CATT | LXMERT† · NLVR2 test-P, Pair | accuracy: 74.4 -> 76.00 | Helped |
| Add CATT to Transformer | Transformer VQA · VQA2.0 test-std Number questions | accuracy: 50.7 -> 53.45 | Helped |
| Add CATT to Transformer | Transformer captioner · MS COCO Karpathy split | CHAIRs: 12.1 -> 9.7 | Helped |
| Add CATT to Transformer | Transformer captioner · MS COCO Karpathy split | gender accuracy: 47% -> 56% | Helped |
| Use K-means dictionary initialization (ablation) | LXMERT+CATT · VQA2.0 local validation | accuracy: 69.81 -> 70.40 | Helped |
| Share IS-ATT and CS-ATT parameters (ablation) | LXMERT+CATT · VQA2.0 local validation | accuracy: 70.05 -> 70.40 | Helped |
Closest papers by meaning
- Rethinking Causal Mask Attention for Vision-Language Inference (ICLR 2026 · similarity 0.90)
- Causal Tracing of Object Representations in Large Vision Language Models: Mechanistic Interpretability and Hallucination Mitigation (AAAI 2026 · similarity 0.88)
- Seeing is Understanding: Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs (ICML 2026 · similarity 0.87)
- Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration (ICML 2024 · similarity 0.87)
- See What You Are Told: Visual Attention Sink in Large Multimodal Models (ICLR 2025 · similarity 0.87)
- Learning 1D Causal Visual Representation with De-focus Attention Networks (NeurIPS 2024 · similarity 0.87)
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.