AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Towards Higher Effective Rank in Parameter-Efficient Fine-tuning using Khatri-Rao Product

Paul Albert, Frederic Z. Zhang, Hemanth Saratchandran, Anton van den Hengel, Ehsan Abbasnejad

ICCV 2025 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 7 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 KRAdapter instead of LoRACLIP ViT-B/32 · VTAB1kaccuracy: 66.1 -> 71.1Improved
Use KRAdapter instead of LoRACLIP ViT-H/14 · VTAB1kaccuracy: 75.9 -> 77.0Improved
Use KRAdapter instead of full fine-tuningCLIP ViT-H/14 · VTAB1kaccuracy: 71.3 -> 77.0Improved
Use KRAdapter instead of LoRA, distribution shiftCLIP ViT-H/14 · ImageNet OOD (7 datasets)OOD accuracy: 75.9 -> 78.8Improved
Use KRAdapter instead of LoRA, generalisation ratioCLIP ViT-H/14 · ImageNet OOD (7 datasets)r_gen: 0.04 -> 0.48Improved
Use KRAdapter instead of LoRALlama3.1-8B 4-bit · Commonsense reasoningavg accuracy: 68.34 -> 72.43Improved
Use KRAdapter instead of LoRAQwen2.5-7B 4-bit · Commonsense reasoningavg accuracy: 69.58 -> 79.28Improved
Use KRAdapter instead of LoRA on OOD setsLlama3.1-8B 4-bit · Commonsense reasoning OODaccuracy: 55.62 -> 64.66Improved
Use KRAdapter instead of Krona, avg accuracyQwen2.5-7B 4-bit · Commonsense reasoningavg accuracy: 78.10 -> 79.28Improved
Use KRAdapter instead of Krona on ID ImageNetCLIP ViT-H/14 · ImageNet IDaccuracy: 85.0 -> 84.7Hurt
Use KRAdapter instead of RandLoRA on structured subsetCLIP ViT-H/14 · VTAB1k structuredaccuracy: 59.5 -> 58.1Hurt
Double trainable parameters beyond the low configuration (ablation)KRAdapter ViT-H/14 · 11 CLIP datasetsaccuracy: no increase beyond low configurationNo clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 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