Research map / Vision-language and multimodal / Multimodal LLMs
Robust Disentangled Counterfactual Learning for Physical Audiovisual Commonsense Reasoning
Mengshi Qi, Changsheng Lv, Huadóng Ma
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 DCL to Late Fusion | Late Fusion · PACS | accuracy (%): 55.0 -> 57.7 | Improved |
| add DCL to CLIP | CLIP · PACS | accuracy (%): 56.3 -> 58.4 | Improved |
| add DCL to UNITER | UNITER · PACS | accuracy (%): 60.6 -> 62.0 | Improved |
| add DCL to AudioCLIP | AudioCLIP · PACS | accuracy (%): 60.0 -> 63.2 | Improved |
| add DSE+ contrastive losses on top of DCL (ablation) | AudioCLIP · PACS | accuracy (%): 63.2 -> 65.3 | Improved |
| add DCL to AudioCLIP | AudioCLIP · PACS-Material | accuracy (%): 75.9 -> 76.2 | Improved |
| replace parameter-matched MLP with DSE (ablation) | AudioCLIP · PACS | accuracy (%): 60.3 -> 61.1 | Improved |
| add MLP instead of DSE to Late Fusion (ablation) | Late Fusion · PACS | accuracy (%): 55.0 -> 54.9 | No clear effect |
| add physical knowledge relationship A without disentanglement (ablation) | CLIP · PACS-Material | accuracy (%): 72.4 -> 51.5 | Hurt |
| add physical knowledge relationship A without disentanglement (ablation) | AudioCLIP · PACS-Material | accuracy (%): 75.9 -> 70.3 | Hurt |
| add DSE before relationship A (ablation) | CLIP · PACS | accuracy (%) | Improved |
| add counterfactual intervention C to relationship A (ablation) | CLIP · PACS-Material | accuracy (%): 51.5 -> 68.9 | Improved |
Closest papers by meaning
- Counterfactual Debiasing for Physical Audiovisual Commonsense Reasoning (AAAI 2025 · similarity 0.91)
- When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion? (AAAI 2026 · similarity 0.85)
- What If the TV Was Off? Examining Counterfactual Reasoning Abilities of Multi-modal Language Models (CVPR 2024 · similarity 0.84)
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.84)
- Evidence-RL: Towards Evidence-intensive Visual Reasoning (NeurIPS 2026 · similarity 0.84)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.83)
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.
