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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add OSP layer to autoencoder (no disentanglement objective) (ablation) | AE · 3D Shapes | MIG: 5.46 -> 8.87 | Improved |
| add OSP layer to beta-TCVAE | beta-TCVAE · 3D Shapes | MIG: 74.70 -> 84.88 | Improved |
| add OSP layer to CLG-VAE | CLG-VAE · 3D Shapes | SAP: 34.41 -> 34.79 | No clear effect |
| add OSP layer to every conv/linear layer | ResNet-18 · ImageNet | top-1 accuracy: 64.44 -> 65.71 | Improved |
| add OSP layer to ViT-B/16 | ViT-Base/16 · ImageNet | top-1 accuracy: 78.22 -> 80.33 | Improved |
| add OSP layer to YOLOv8-large-seg | YOLOv8-large-seg · MSCOCO | box mAP50-95: 51.28 -> 51.43 | Improved |
| add OSP layers to BERT | BERT · Word analogy | average score (pmi): 34.82 -> 34.42 | Hurt |
| add OSP layers to BERT | BERT · Word analogy | average score (mppl): 33.32 to 34.73 | Improved |
| add OSP layer to BERT | BERT · GLUE CoLA | MCC: 53.91 -> 53.39 | Hurt |
| add OSP layer to BERT-Large | BERT-Large · GLUE RTE | accuracy: 63.54 -> 94.62 | Improved |
| use OSP-fine-tuning instead of full fine-tuning | ResNet-18 · ImageNet | top-1 accuracy: 66.69 -> 67.78 | Improved |
| use OSP-fine-tuning instead of full fine-tuning | ResNet-50 · ImageNet | top-1 accuracy: 76.59 -> 77.67 | Improved |
Closest papers by meaning
- Disentangling Factors of Variation with Cycle-Consistent Variational Auto-Encoders (ECCV 2018 · similarity 0.88)
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.87)
- NashAE: Disentangling Representations through Adversarial Covariance Minimization (ECCV 2022 · similarity 0.87)
- Towards Principled Disentanglement for Domain Generalization (CVPR 2022 · similarity 0.87)
- Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations (NeurIPS 2025 · similarity 0.86)
- Can Diffusion Models Disentangle? A Theoretical Perspective (NeurIPS 2025 · similarity 0.85)
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.
