AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Context-Aware Robust Fine-Tuning

Xiaofeng Mao, Yufeng Chen, Xiaojun Jia, Rong Zhang, Hui Xue, Li Zhao

IJCV 2023 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 27 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 CAR-FT context KL loss to fine-tuningCLIP ViT-B/16 FT · ImageNet distribution shiftsavg top-1 accuracy: 45.1 -> 56.9Improved
use CAR-FT instead of LP-FTCLIP ViT-B/16 · ImageNet valtop-1 accuracy: 81.7 -> 83.3Improved
combine CAR-FT with WiSE-FTWiSE-FT · ImageNet distribution shiftsavg top-1 accuracy: 58.1 -> 58.9Improved
apply WiSE-FT to CAR-FTCAR-FT · ImageNet valtop-1 accuracy: 83.3 -> 82.1Hurt
add CAR-FT to greedy model soupsGreedy Model Soups · ImageNet distribution shiftsavg top-1 accuracy: 56.3 -> 57.1Improved
add CAR-FT to uniform model soupsUniform Model Soups · ImageNet distribution shiftsavg top-1 accuracy: 57.5 -> 58.2Improved
use CAR-FT instead of ERMCLIP ViT-B/16 · DomainBedavg accuracy: 72.2 -> 78.5Improved
use CAR-FT instead of WiSE-FTCLIP ViT-B/16 · Nine downstream tasksavg accuracy: 83.8 -> 84.5Improved
use text-prompt head init (TP-FT) instead of random init (FT)CLIP ViT-B/16 · Flowers102accuracy: 93.7 -> 93.1Hurt
add CAR-FT on small-data CLIP variantsSLIP-YFCC15M ViT-B/32 · ImageNet distribution shiftsavg top-1 accuracy: 20.7 -> 21.0No clear effect
reduce context templates from 80 to 7 (CLIP templates) (ablation)CAR-FT · ImageNet distribution shiftsavg top-1 accuracy: 56.9 -> 56.5Hurt
use 7 searched templates instead of 7 CLIP templates (ablation)CAR-FT · ImageNet distribution shiftsavg top-1 accuracy: 56.5 -> 56.9Improved

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