Research map / Continual learning / Online and replay-based
VQACL: A Novel Visual Question Answering Continual Learning Setting
Xi Zhang, Feifei Zhang, Changsheng Xu
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 sample-specific (SS) and sample-invariant (SI) representation learning with rehearsal | transformer encoder-decoder VQA continual learner · VQA v2 standard testing | AP (%): +1.78 to +4.74 over rehearsal baselines (ER, DER, VS) | Improved |
| add SS+SI representation learning with rehearsal | transformer encoder-decoder VQA continual learner · NExT-QA standard testing | AP (WUPS): +1.72 to +6.1 over rehearsal baselines | Improved |
| add SS+SI representation learning with rehearsal | transformer encoder-decoder VQA continual learner · VQA v2 and NExT-QA novel composition testing | AP (%): +1.62 to +22.57 (VQA v2); +2.02 to +22.31 (NExT-QA) over baselines | Improved |
| remove sample-specific feature (ablation) | full model · VQA v2 Group-1 novel composition | AP (%): 35.50 -> 15.49 | Hurt |
| add sample-invariant (SI) feature (ablation) | full model · VQA v2 Group-1 novel composition | AP (%): 31.86 -> 35.50 | Improved |
| use separate question and visual SI features instead of single fused SI feature (ablation) | full model · VQA v2 Group-1 novel composition | AP (%): 32.35 -> 35.50 | Improved |
| use separate question and visual SI features instead of single fused SI feature (ablation) | full model · VQA v2 standard testing | AP (%): 31.88 -> 32.27 | Improved |
| add sample-invariant (SI) feature (ablation) | full model · VQA v2 standard testing | AP (%): 30.55 -> 32.27 | Improved |
| switch regularization (EWC, MAS) to rehearsal methods | transformer encoder-decoder baseline · VQA v2 standard testing | AP (%): EWC 15.77 / MAS 20.56 vs ER 36.99 | Improved |
| add DER and VS rehearsal instead of ER | transformer encoder-decoder baseline · NExT-QA standard testing | AP (WUPS): ER 30.55 vs DER 26.17, VS 28.13 | Hurt |
| increase memory size from 1000 to 5000 (VQA v2) or 100 to 500 (NExT-QA) | rehearsal methods · VQA v2 and NExT-QA standard and composition testing | AP (%) | Mixed |
| change linguistic task order (CW, DL, TN first) | Vanilla model · NExT-QA | AP (%): 0.42 to 2.19 variation in last-task AP | No clear effect |
Closest papers by meaning
- Re-evaluating Continual VQA: Toward Fair and Robust Evaluation for Multimodal Continual Learning (CVPR 2026 · similarity 0.95)
- Ask and Remember: A Questions-Only Replay Strategy for Continual Visual Question Answering (ICCV 2025 · similarity 0.94)
- Symbolic Replay: Scene Graph as Prompt for Continual Learning on VQA Task (AAAI 2023 · similarity 0.94)
- AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning (CVPR 2025 · similarity 0.93)
- CL-MoE: Enhancing Multimodal Large Language Model with Dual Momentum Mixture-of-Experts for Continual Visual Question Answering (CVPR 2025 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
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.
