AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

OPLoRA: Orthogonal Projection LoRA Prevents Catastrophic Forgetting During Parameter-Efficient Fine-Tuning

Yifeng Xiong, Xiaohui Xie

AAAI 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 14 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 LoRA with OPLoRA-128LLaMA-2 7B · commonsense avg (fine-tune + forgetting)average accuracy: 57.03 -> 57.76Improved
replace LoRA with OPLoRA-16Qwen2.5 7B · commonsense avg (fine-tune + forgetting)average accuracy: 63.24 -> 64.26Improved
replace LoRA with OPLoRA-128 (forgetting MBPP)LLaMA-2 7B · MBPP after commonsense fine-tuningaccuracy: 20.2 -> 20.4Improved
replace LoRA-Null with OPLoRA-128 (forgetting MBPP)LLaMA-2 7B · MBPP after commonsense fine-tuningaccuracy: 9.8 -> 20.4Improved
replace LoRA with OPLoRA-16LLaMA-2 7B · GSM8Kaccuracy: 44.88 -> 49.73Improved
replace LoRA with OPLoRA-16Qwen2.5 7B · GSM8Kaccuracy: 82.18 -> 83.70Improved
replace LoRA with OPLoRA-16LLaMA-2 7B · MBPPpass rate (%): 37.6 -> 40.5Improved
replace LoRA with OPLoRA-16LLaMA-2 7B · MBPP++pass rate (%): 30.4 -> 34.1Improved
replace LoRA with OPLoRA-16Qwen2.5 7B · MBPPpass rate (%): 80.2 -> 80.2No clear effect
use PiSSA instead of LoRALLaMA-2 7B · commonsense avg (fine-tune + forgetting)average accuracy: 57.03 -> 32.42Hurt
use projection rank k=128 instead of k=16OPLoRA · tasks overallaccuracyMixed
use OPLoRA instead of LoRA (training cost)LLaMA-2 7B commonsense · training timehours: 5 h 13 m -> 6 h 12 mHurt

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