Research map / Vision-language and multimodal / Visual question answering
Greedy Gradient Ensemble for Robust Visual Question Answering
Xinzhe Han, Shuhui Wang, Chi Su, Qingming Huang, Qi Tian
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 greedy ensemble instead of summing biased outputs (SUM-DQ) (ablation) | GGE-DQ · VQA-CP v2 test | accuracy (All): 35.46 -> 57.12 | Improved |
| use greedy ensemble instead of LMH+RUBi combination (ablation) | GGE-DQ · VQA-CP v2 test | accuracy (All): 51.54 -> 57.12 | Improved |
| add question-shortcut bias to distribution-only GGE-D (ablation) | GGE · VQA-CP v2 test | accuracy (All): 48.27 -> 57.12 | Improved |
| add distribution bias to question-only GGE-Q-iter (ablation) | GGE · VQA-CP v2 test | accuracy (All): 43.72 -> 57.12 | Improved |
| add GGE-DQ-tog to base model S-MRL (ablation) | S-MRL · VQA-CP v2 test | accuracy (All): 37.90 -> 54.62 | Improved |
| add GGE-DQ-tog to base model BAN (ablation) | BAN · VQA-CP v2 test | accuracy (All): 35.94 -> 51.91 | Improved |
| use self-ensemble GGE-SF-iter without predefined bias (ablation) | UpDn · VQA-CP v2 test | accuracy (All): 39.89 -> 44.53 | Improved |
| use softmax+CE loss in GGE-DQ-iter instead of sigmoid+BCE (ablation) | GGE-DQ-iter · VQA-CP v2 test | accuracy (All): 57.12 -> 56.25 | Hurt |
| invert human grounding annotations in HINT (ablation) | HINT · VQA-CP | accuracy (All): 47.50 -> 47.20 | No clear effect |
| add inverse supervision to UpDn (ablation) | UpDn · VQA-CP | accuracy (All): 39.89 -> 42.12 | Improved |
| add inverse supervision to LMH | LMH · VQA v2 val | accuracy (All): 56.35 -> 43.29 | Hurt |
| add GGE-DQ-tog to UpDn on in-distribution data | UpDn · VQA v2 val | accuracy (All): 63.79 -> 59.11 | Hurt |
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.
- RUBi — RUBi: Reducing Unimodal Biases for Visual Question Answering (NeurIPS 2019) · 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
- Language‑Bias‑Resilient Visual Question Answering via Adaptive Multi‑Margin Collaborative Debiasing (NeurIPS 2025 · similarity 0.94)
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.94)
- Generative Bias for Robust Visual Question Answering (CVPR 2023 · similarity 0.94)
- Robust Visual Question Answering: Datasets, Methods, and Future Challenges (TPAMI 2024 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.89)
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.
