AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Layer-Centric Factors of Variation Disentanglement for Task- and Model-Agnostic Generalization

Hee-Jun Jung, Jongmin Park, Minwoo Kang, Hoyong Kim, Kangil Kim

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging

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 OSP layer to autoencoder (no disentanglement objective) (ablation)AE · 3D ShapesMIG: 5.46 -> 8.87Improved
add OSP layer to beta-TCVAEbeta-TCVAE · 3D ShapesMIG: 74.70 -> 84.88Improved
add OSP layer to CLG-VAECLG-VAE · 3D ShapesSAP: 34.41 -> 34.79No clear effect
add OSP layer to every conv/linear layerResNet-18 · ImageNettop-1 accuracy: 64.44 -> 65.71Improved
add OSP layer to ViT-B/16ViT-Base/16 · ImageNettop-1 accuracy: 78.22 -> 80.33Improved
add OSP layer to YOLOv8-large-segYOLOv8-large-seg · MSCOCObox mAP50-95: 51.28 -> 51.43Improved
add OSP layers to BERTBERT · Word analogyaverage score (pmi): 34.82 -> 34.42Hurt
add OSP layers to BERTBERT · Word analogyaverage score (mppl): 33.32 to 34.73Improved
add OSP layer to BERTBERT · GLUE CoLAMCC: 53.91 -> 53.39Hurt
add OSP layer to BERT-LargeBERT-Large · GLUE RTEaccuracy: 63.54 -> 94.62Improved
use OSP-fine-tuning instead of full fine-tuningResNet-18 · ImageNettop-1 accuracy: 66.69 -> 67.78Improved
use OSP-fine-tuning instead of full fine-tuningResNet-50 · ImageNettop-1 accuracy: 76.59 -> 77.67Improved

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