Research map / Video / Video-language understanding
Knowledge Proxy Intervention for Deconfounded Video Question Answering
Jiangtong Li, Li Niu, Liqing Zhang
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 KPI front-door adjustment to baselines | HQGA, HGA, CoMem · five VideoQA datasets | accuracy: +1.4% to +5.3% | Improved |
| add KPI to HQGA on MSVD-QA | HQGA · MSVD-QA | accuracy: 41.2 -> 43.3 | Improved |
| add KPI to pretrained VIOLET | VIOLET · Causal-VidQA | accuracy: 58.6 -> 59.3 | Improved |
| use multi-head attention instead of channel or product attention in EXP (ablation) | KPI with HQGA · MSVD-QA | accuracy: 41.5 -> 43.3 | Improved |
| remove the knowledge proxy Z from KPI (ablation) | KPI with HGA · MSVD-QA | accuracy: 41.2 -> 36.8 | Hurt |
| remove aligned feature H expectation (ablation) | KPI with HGA · NExT-QA | accuracy: 54.1 -> 53.2 | Hurt |
| use both ConceptNet and Atomic for knowledge space (ablation) | KPI with HQGA · NExT-QA | accuracy: 54.6 -> 55.5 | Improved |
| share knowledge space across MSVD-QA and NExT-QA (ablation) | KPI · MSVD-QA vs NExT-QA | accuracy: MSVD-QA 43.3 -> 41.9; NExT-QA 55.0 -> 55.5 (HQGA) | Mixed |
| increase feature space size from 100 to 1000 (ablation) | KPI · MSVD-QA, NExT-QA | accuracy | 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.
- MERLOT — MERLOT: Multimodal Neural Script Knowledge Models (NeurIPS 2021) · better on 1 of 1 shared result
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
- Discovering the Real Association: Multimodal Causal Reasoning in Video Question Answering (CVPR 2023 · similarity 0.90)
- Two Causally Related Needles in a Video Haystack (NeurIPS 2025 · similarity 0.88)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.87)
- Cross-modal Causal Relation Alignment for Video Question Grounding (CVPR 2025 · similarity 0.87)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.87)
- Align and Aggregate: Compositional Reasoning with Video Alignment and Answer Aggregation for Video Question-Answering (CVPR 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.
