AI Paper Atlas

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

TPAMI 2025 · Continual learning › Continual fine-tuning of LLMs · approach: Mixture-of-experts for multimodal continual tuning · 115 citations (OpenAlex)

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 changedWhereMeasuredEffect
add expandable projections to CLIP (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
add cross-modal fusion to CLIP (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
combine projection and fusion (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
add context prompts on top of projection and fusion (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
vary context prompt length c from 1 to 100 (ablation)PROOF · CIFAR100 B0 Inc10accuracyNo clear effect
use a single linear layer instead of SSF or Adapter projection (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
use visual prototypes, textual prototypes and context prompts together as context (ablation)PROOF · CIFAR100 B0 Inc10accuracyImproved
merge projections into a single one at inference (ablation)PROOF · CIFAR100 B0 Inc10extra parameters: (2b+3)xd^2 + Bxd -> 5xd^2 + BxdImproved
use residual projection output (PROOF dagger)PROOF · CIFAR100 B0 Inc10zero-shot accuracy on unseen classesMixed
switch from LAION-400M to OpenAI CLIP weights (ablation)PROOF · nine CIL benchmarksaccuracyNo clear effect
use PROOF over CODA-PromptCLIP ViT-B/16 · Aircraft B0 Inc10average accuracy: 63.13 -> 64.61Improved
use PROOF over FinetuneCLIP ViT-B/16 · Flickr30K image to textR@1 last stage: 48.79 -> 85.68Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy