AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

L2-LoRA: Improving Low-Rank Adaptation with Layer-Specific Regularization

Xiang Zhang, Rui Xie, Shikun Zhang

AAAI 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 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 layer-specific L2 regularization to LoRALoRA (Llama-7B) · MMLU avg after Commonsense170Kaccuracy: 23.1 -> 28.3Improved
add layer-specific L2 regularization to LoRALoRA (Llama-7B) · ARC-caccuracy: 61.3 -> 65.6Improved
add layer-specific L2 regularization in continual learningLoRA (Llama-7B) · MMLU avg, stage 1accuracy: 25.7 -> 28.3Improved
add layer-specific L2 regularization in continual learningLoRA (Llama-7B) · MMLU avg, stage 3accuracy: 23.3 -> 23.6No clear effect
add layer-specific L2 regularizationLoRA (Llama2-7B) · PIQAaccuracy: 79.9 -> 85.2Improved
add layer-specific L2 regularizationLoRA (Llama2-7B) · BoolQaccuracy: 69.8 -> 68.9Hurt
add layer-specific L2 regularizationLoRA (Llama2-7B) · GSM8Kaccuracy: 60.58 -> 65.32Improved
add layer-specific L2 regularizationLoRA (Llama2-7B) · MATHaccuracy: 16.88 -> 18.12Improved
fine-tune only lowest layers vs highest layers (ablation)LoRA (Llama2-7B) · MMLUaccuracy: 43.95 -> 41.71 (low, 6 layers); 43.88 (high)Hurt
fine-tune highest layers instead of lowest layers (ablation)LoRA (Llama2-7B) · AddSubaccuracy: 11.14 -> 17.22Improved

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