AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

StolenLoRA: Exploring LoRA Extraction Attacks via Synthetic Data

Yixu Wang, Yan Teng, Yingchun Wang, Xingjun Ma

ICCV 2025 · Adversarial robustness and security › Backdoor and poisoning · approach: Detecting poisoned samples · 0 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 structured prompt template for LLM (ablation)StolenLoRA-Rand · CUBS200 IBaccuracy: 72.33 -> 75.35Improved
add structured prompt template for LLM (ablation)StolenLoRA-Rand · CUBS200 XBaccuracy: 42.21 -> 45.70Improved
add LLM prompt generation (ablation)StolenLoRA-Rand · Indoor67 XBaccuracy: 47.09 -> 59.18Improved
add LLM prompt generation (ablation)StolenLoRA-Rand · CUBS200 IBaccuracy: 70.66 -> 75.35Improved
use GPT-4o instead of GPT-4o mini (ablation)StolenLoRA-Rand · Indoor67 XBaccuracy: 59.18 -> 63.58Improved
add DSL querying instead of random querying (ablation)StolenLoRA · CUBS200 XBaccuracy: 45.70 -> 50.14Improved
add DSL querying instead of random queryingStolenLoRA · CUBS200 IBaccuracy: 75.35 -> 73.23Hurt
add label refining loss L_LR (ablation)StolenLoRA-DSL · CUBS200 IBaccuracy: 70.17 -> 73.23Improved
add label refining loss L_LR (ablation)StolenLoRA-DSL · Indoor67 XBaccuracy: 64.10 -> 65.07Improved
raise DSL confidence threshold tau from 0.5 to 0.95 (ablation)StolenLoRA-DSL · CUBS200 XBaccuracy: 48.43 -> 50.14Improved
raise DSL confidence threshold tau from 0.95 to 0.99 (ablation)StolenLoRA-DSL · Indoor67 XBaccuracy: 65.07 -> 61.72Hurt
use hard labels instead of soft labels (ablation)StolenLoRA-Rand · CUBS200 IBaccuracy: 75.35 -> 67.00Hurt

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