Research map / Continual learning / Continual fine-tuning of LLMs
Learning Without Forgetting for Vision-Language Models
Da-Wei Zhou, Yuanhan Zhang, Yan Wang, Jingyi Ning, Han-Jia Ye, De‐Chuan Zhan, Ziwei Liu
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 expandable projections to CLIP (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| add cross-modal fusion to CLIP (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| combine projection and fusion (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| add context prompts on top of projection and fusion (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| vary context prompt length c from 1 to 100 (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | No clear effect |
| use a single linear layer instead of SSF or Adapter projection (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| use visual prototypes, textual prototypes and context prompts together as context (ablation) | PROOF · CIFAR100 B0 Inc10 | accuracy | Improved |
| merge projections into a single one at inference (ablation) | PROOF · CIFAR100 B0 Inc10 | extra parameters: (2b+3)xd^2 + Bxd -> 5xd^2 + Bxd | Improved |
| use residual projection output (PROOF dagger) | PROOF · CIFAR100 B0 Inc10 | zero-shot accuracy on unseen classes | Mixed |
| switch from LAION-400M to OpenAI CLIP weights (ablation) | PROOF · nine CIL benchmarks | accuracy | No clear effect |
| use PROOF over CODA-Prompt | CLIP ViT-B/16 · Aircraft B0 Inc10 | average accuracy: 63.13 -> 64.61 | Improved |
| use PROOF over Finetune | CLIP ViT-B/16 · Flickr30K image to text | R@1 last stage: 48.79 -> 85.68 | Improved |
Closest papers by meaning
- Don't Forget Why You Started: Tackling Dual Forgetting in Vision-Language Continual Learning (ICML 2026 · similarity 0.92)
- Enhanced Continual Learning of Vision-Language Models with Model Fusion (ICLR 2026 · similarity 0.92)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.89)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.89)
- Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters (CVPR 2024 · similarity 0.88)
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.
