AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Geometry-Preserving Orthonormal Initialization for Low-Rank Adaptation in RLVR

Ruijia Zhang, Jiacheng Zhu, Hanqing Zhu, Laixi Shi

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Low-rank adaptation variants · 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
use orthonormal principal-subspace init without singular-value scaling (LoRA-RLPO) instead of PiSSADeepSeek-R1-Distill-Qwen-1.5B LoRA RLVR · Average of 5 benchmarks step 500accuracy (%): 4.57 -> 65.03Improved
use LoRA-RLPO instead of standard LoRADeepSeek-R1-Distill-Qwen-1.5B LoRA RLVR · Average of 5 benchmarks step 500accuracy (%): 62.40 -> 65.03Improved
use LoRA-RLMO (orthonormal minor-subspace init) instead of MiLoRADeepSeek-R1-Distill-Qwen-1.5B LoRA RLVR · Average of 5 benchmarks step 500accuracy (%): 51.71 -> 63.76Improved
use LoRA-RLMO instead of standard LoRADeepSeek-R1-Distill-Qwen-1.5B LoRA RLVR · Average of 5 benchmarks step 500accuracy (%): 62.40 -> 63.76Improved
use LoRA-RLPO instead of standard LoRALoRA RLVR · MATH500 (mean@4) step 500accuracy (%): 81.93 -> 87.33Improved
use LoRA-RLMO instead of standard LoRALoRA RLVR · AIME22 (mean@32) step 500accuracy (%): 43.33 -> 42.22No clear effect
use LoRA-RLMO instead of standard LoRALoRA RLVR · GSM8K (mean@1) step 500accuracy (%): 75.64 -> 76.42No clear effect
add orthonormal data-agnostic init (DCT-LoRA, Wavelet-LoRA) to LoRA (ablation)LoRA RLVR · DeepSeek-R1-Distill-Qwen-1.5B five math benchmarksaccuracyImproved
add principal-subspace SVD geometry on top of orthonormality (LoRA-RLPO vs DCT/Wavelet-LoRA) (ablation)LoRA RLVR · average accuracyaccuracyImproved
add minor-subspace SVD geometry on top of orthonormality (LoRA-RLMO vs DCT/Wavelet-LoRA) (ablation)LoRA RLVR · average accuracyaccuracyNo clear effect
decay learning rate for MiLoRA (ablation)MiLoRA RLVR · training stabilitytraining stabilityImproved
decay learning rate for PiSSA (ablation)PiSSA RLVR · training stabilitytraining stabilityNo 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