AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Unlocking Pre-trained Weights: Parameter Inheritance for Zero-Shot Initialization

Jiaze Xu, Shiyu Xia, Jiaqi Lv, Xin Geng

CVPR 2026 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Robust fine-tuning of zero-shot models

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
inherit pre-trained ViT-Large weights through learned projection (PITH) instead of soft-label supervision (TAL)ViT-Base hypernetwork initialization · ImageNet-1K zero-shottop-1 accuracy (%): 46.81 -> 53.35Improved
add pre-trained parameter projection (ablation)PITH · ImageNet-1K untrained averageaverage accuracy (%): 46.90 -> 49.02Improved
add progressive dimension expansion (ablation)PITH · ImageNet-1K untrained averageaverage accuracy (%): 46.43 -> 49.02Improved
share MLP layers between weight and projection pathways (ablation)PITH · ImageNet-1K untrained averageaverage accuracy (%): 45.73 -> 46.08 vs LoGAH, params 21.41M -> 24.40MImproved
use full dual-pathway decoder instead of shared pathway (ablation)PITH · ImageNet-1K untrained averageaverage accuracy (%): 46.08 -> 46.43 with params 24.40M -> 41.59MImproved
PITH instead of TAL trained longer (300 epochs)hypernetwork training budget · Decathlon average accuracyaverage accuracy (%): 47.54 -> 51.03Improved
PITH initialization instead of TALViT-Tiny · Decathlon trained 100 epochsaverage accuracy (%): 58.98 -> 61.65Improved
PITH initialization instead of TALViT-Small · Decathlon trained 100 epochsaverage accuracy (%): 58.39 -> 60.73Improved
PITH initialization instead of TALViT-Tiny 12 layers · Decathlon untrained, DPedaccuracy (%): 93.11 -> 90.56Hurt
PITH initialization instead of TALViT-Small · FER2013 unseen task, 20 epochsaccuracy (%): 53.50 -> 52.55Hurt
PITH initialization instead of RandInitViT-Tiny · HAM10000 unseen task, 100 epochsaccuracy (%): 97.34 -> 97.71No clear effect
PITH instead of TALtraining cost · 200 epochs on RTX Pro 6000training time (h): 92.37 -> 95.43No clear effect

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