Research map / Continual learning / Online and replay-based
Symbolic Replay: Scene Graph as Prompt for Continual Learning on VQA Task
Stan Weixian Lei, Difei Gao, Jay Zhangjie Wu, Yuxuan Wang, Wei Liu, Mengmi Zhang, Mike Zheng Shou
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 symbolic replay of scene graphs and associated QA pairs instead of no replay | UniVQA on CLOVE-scene · CLOVE-scene (six task orders) | average accuracy (%): 26.92 -> 33.05 | Improved |
| Use symbolic replay of scene graphs and associated QA pairs instead of no replay | UniVQA on CLOVE-function · CLOVE-function (six task orders) | average accuracy (%): 26.16 -> 42.01 | Improved |
| Use symbolic replay rather than VQG image-and-answer replay | UniVQA on CLOVE-scene · CLOVE-scene (six task orders) | average accuracy (%): 29.19 -> 33.05 | Improved |
| Use symbolic replay rather than VQG image-and-answer replay | UniVQA on CLOVE-function · CLOVE-function (six task orders) | average accuracy (%): 31.36 -> 42.01 | Improved |
| Replace random SG prompts with stored ground-truth SG prompts (ablation) | SGP · CLOVE-scene (γ=0.9) | average accuracy (%): 32.08 -> 35.09 | Improved |
| Replace random SG prompts with stored ground-truth SG prompts (ablation) | SGP · CLOVE-function (γ=0.9) | average accuracy (%): 44.21 -> 47.01 | Improved |
| Add scene-graph replay to QA-pair replay (ablation) | SGP · CLOVE-scene (γ=0.9) | average accuracy (%): 29.52 -> 32.08 | Improved |
| Add scene-graph replay to QA-pair replay (ablation) | SGP · CLOVE-function (γ=0.9) | average accuracy (%): 40.24 -> 44.21 | Improved |
| Use symbolic scene-graph replay instead of stored real-data replay | UniVQA on CLOVE-scene · CLOVE-scene (six task orders) | average accuracy (%): 37.19 -> 33.05 | Hurt |
| Use symbolic scene-graph replay instead of stored real-data replay | UniVQA on CLOVE-function · CLOVE-function (six task orders) | average accuracy (%): 41.84 -> 42.01 | No clear effect |
| Train SRM with only 1% of scene-graph annotations (ablation) | SGP on CLOVE-function-oarlks · CLOVE-function | average accuracy (%): +4.9% | Improved |
| Increase replay amount beyond γ=0.7 (ablation) | UniVQA on CLOVE-function · CLOVE-function | average accuracy | No clear effect |
Closest papers by meaning
- VQACL: A Novel Visual Question Answering Continual Learning Setting (CVPR 2023 · similarity 0.94)
- Ask and Remember: A Questions-Only Replay Strategy for Continual Visual Question Answering (ICCV 2025 · similarity 0.92)
- Re-evaluating Continual VQA: Toward Fair and Robust Evaluation for Multimodal Continual Learning (CVPR 2026 · similarity 0.91)
- MacVQA: Adaptive Memory Allocation and Global Noise Filtering for Continual Visual Question Answering (AAAI 2026 · similarity 0.91)
- AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning (CVPR 2025 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 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.
