Research map / Continual learning / Continual fine-tuning of LLMs
SCoT: Similarity-guided Conflict-aware Task Consolidation for Continual VQA
Anand Patel, Moloud Abdar, Biplab Banerjee
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 similarity-guided layer-wise consolidation (C2) to uniform task-vector averaging (ablation) | SCoT · VQAv2 standard test | AP (%): 39.23 -> 42.62 | Improved |
| add similarity-guided layer-wise consolidation (C2) (ablation) | SCoT · NExT-QA standard test | AP (%): 27.35 -> 32.15 | Improved |
| add similarity-guided consolidation (C2) (ablation) | SCoT · VQAv2 standard test | AF (%): -0.35 -> 1.42 | Hurt |
| add conflict-aware projection (C3) (ablation) | SCoT · VQAv2 standard test | AF (%): 1.42 -> 0.92 | Improved |
| add conflict-aware projection (C3) (ablation) | SCoT · NExT-QA novel composition test | AP (%): 34.00 -> 33.75 | Hurt |
| add depth-aware stabilization (C4) (ablation) | SCoT · VQAv2 novel composition test | AP (%): 41.31 -> 42.11 | Improved |
| add depth-aware stabilization (C4) (ablation) | SCoT · VQAv2 standard test | AP (%): 43.02 -> 42.78 | Hurt |
| stabilize late encoder and early decoder instead of early encoder layers (ablation) | SCoT · VQAv2 novel composition test | AP (%): 40.38 -> 42.11 | Improved |
| use consolidation range [0.08, 0.45] instead of narrow [0.05, 0.30] (ablation) | SCoT · VQAv2 standard test | AP (%): 41.22 -> 42.78 | Improved |
| use SCoT instead of joint multitask training | SCoT · VQAv2 standard test | AP (%): 41.32 -> 42.78 | Improved |
| use SCoT instead of joint multitask training | SCoT · NExT-QA standard test | AP (%): 35.92 -> 32.28 | Hurt |
| remove replay buffer (ablation) | SCoT · VQAv2 standard test | AP (%): 42.78 -> 31.44 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MagMax — MagMax: Leveraging Model Merging for Seamless Continual Learning (ECCV 2024) · better on 0 of 2 shared results
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
- Re-evaluating Continual VQA: Toward Fair and Robust Evaluation for Multimodal Continual Learning (CVPR 2026 · similarity 0.94)
- VQACL: A Novel Visual Question Answering Continual Learning Setting (CVPR 2023 · similarity 0.93)
- Ask and Remember: A Questions-Only Replay Strategy for Continual Visual Question Answering (ICCV 2025 · similarity 0.93)
- MacVQA: Adaptive Memory Allocation and Global Noise Filtering for Continual Visual Question Answering (AAAI 2026 · similarity 0.93)
- AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning (CVPR 2025 · similarity 0.90)
- CL-MoE: Enhancing Multimodal Large Language Model with Dual Momentum Mixture-of-Experts for Continual 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.
