Research map / Continual learning / Continual fine-tuning of LLMs
Mind the Interference: Retaining Pre-trained Knowledge in Parameter Efficient Continual Learning of Vision-Language Models
Longxiang Tang, Zhuotao Tian, Kai Li, Chunming He, Hantao Zhou, Hengshuang Zhao, Xiu Li, Jiaya Jia
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 zero-initialization to prompt tuning (without residual attention) (ablation) | prompt tuning · MTIL | Transfer: 57.7 -> 57.3 | No clear effect |
| replace prompt prepending with residual attention (ablation) | DIKI · MTIL | Transfer: 57.7 -> 59.9 | Improved |
| add zero-initialization to residual attention (ablation) | DIKI · MTIL | Transfer: 59.9 -> 63.1 | Improved |
| add distribution-aware calibration (ablation) | DIKI · MTIL | Transfer: 63.1 -> 68.7 | Improved |
| add zero-init and calibration (ablation) | DIKI · MTIL | Last: 85.2 -> 85.1 | No clear effect |
| increase uniform initialization bound of residual parameters | DIKI · MTIL | Transfer | Hurt |
| DIKI vs ZSCL with 0.86% trainable params and no extra data | CLIP ViT-B/16 · MTIL | Last: 83.6 -> 85.1 | Improved |
| DIKI vs prompt methods on zero-shot retention | CLIP ViT-B/16 · MTIL | Transfer: 52.2 -> 68.7 | Improved |
| replace prompt prepending with IKI in DualPrompt | DualPrompt · 10-split CIFAR-100 | Avg. Acc: 86.51 -> 88.77 | Improved |
| replace prompt prepending with IKI in L2P | L2P · 10-split CIFAR-100 | Forgetting: 7.35 -> 7.28 | No clear effect |
| use fixed larger calibration weight instead of sample-adaptive weight | DIKI · MTIL first task Aircraft | zero-shot ability vs trained-task accuracy | Mixed |
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.
- DualPrompt — DualPrompt: Complementary Prompting for Rehearsal-Free Continual Learning (ECCV 2022) · better on 2 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
- Interference-Aware Continual Vision–Language Learning via Instance-Level Expert Routing (ECCV 2026 · similarity 0.92)
- Enhanced Continual Learning of Vision-Language Models with Model Fusion (ICLR 2026 · similarity 0.91)
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.90)
- Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters (CVPR 2024 · similarity 0.90)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Unified Knowledge Maintenance Pruning and Progressive Recovery with Weight Recalling for Large Vision-Language Models (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.
