AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

ConsNoTrainLoRA: Data-driven Weight Initialization of Low-rank Adapters using Constraints

Debasmit Das, Hyoungwoo Park, Munawar Hayat, Seokeon Choi, Sungrack Yun, Fatih Porikli

ICCV 2025 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 3 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
replace random LoRA init with CNTLoRA-X initLoRA · DreamboothDINO: 62.74 -> 64.63Improved
add variable adapter structure (VAS) to CNTLoRA-X (ablation)CNTLoRA-X · DreamboothDINO: 64.63 -> 65.73Improved
add VAS to CNTLoRA-X (ablation)CNTLoRA-X · DreamboothCLIP-T: 25.83 -> 25.63Hurt
add VAS to CNTLoRA-S (ablation)CNTLoRA-S · myVLMRecall: 0.7390 -> 0.7495Improved
add VAS to CNTLoRA-Sh (ablation)CNTLoRA-Sh · Amazon Product DescriptionSentSim: 0.8273 -> 0.8324Improved
use CNTLoRA-X init with fewer init samples (n=1 vs all) (ablation)CNTLoRA-X · DreamboothDINO: 64.02 -> 64.63No clear effect
replace LoRA with CNTLoRA-X at rank 64LoRA · Dreambooth rank 64DINO: 59.07 -> 62.98Improved
replace LoRA with CNTLoRA-X at rank 32LoRA · Dreambooth rank 32DINO: 56.62 -> 59.09Improved
increase learning rate from 1e-4 to 1e-3 (ablation)CNTLoRA-X · DreamboothDINO: 64.63 -> 61.98Hurt
decrease learning rate from 1e-4 to 1e-5 (ablation)CNTLoRA-X · DreamboothDINO: 64.63 -> 60.05Hurt
use QR decomposition instead of SVD for CNTLoRA-X (ablation)CNTLoRA-X · DreamboothDINO: 64.63 -> 65.29Improved
raise fractional singular allocation p from 0.5 to 0.75 (ablation)CNTLoRA-X · DreamboothDINO / CLIP-T: 64.63 -> 65.78Mixed

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.

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

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