Research map / Label-efficient and robust learning / Domain adaptation and generalization
Adaptive Bayesian Early-Exit Networks for Efficient Non-Transferable Learning
Siyu Luan, Yan Li, Zhong Chen, Zhenyi Wang
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 |
|---|---|---|---|
| set target-loss weight β=1 instead of β=0 (no target term) (ablation) | ENL-DEE · Office-Home, Real-World source (ViT) | PG: 0.00 -> 16.94 | Improved |
| raise β from 1 to 2 (ablation) | ENL-DEE · Office-Home, Real-World source (ViT) | PG: 16.94 -> 9.94 | Hurt |
| raise β from 1 to 3 (ablation) | ENL-DEE · Office-Home, Real-World source (ViT) | PG: 16.94 -> -1.99 | Hurt |
| use two-linear-layer exit heads at blocks 3,6,9,11 vs three-layer heads (ablation) | ENL-DEE · PACS, sketch source (ViT) | PG: 23.59 vs 19.96 | No clear effect |
| move exit placement to blocks 2,5,8,10 (ablation) | ENL-DEE · PACS, sketch source (ViT) | PG: 23.59 -> 19.98 | No clear effect |
| use ENL-DEE instead of NTL with ViT | ENL-DEE · CIFAR-10 source (ViT) | PG: 44.9 -> 50.5 | Improved |
| use ENL-DEE instead of NTL with ResNet-34 | ENL-DEE · CIFAR-10 source (ResNet-34) | PG: 54.9 -> 51.7 | Hurt |
| use ENL-DEE instead of CUTI with ResNet-34 | ENL-DEE · STL-10 source (ResNet-34) | PG: 13.7 -> 38.9 | Improved |
| use ENL-DEE instead of NTL on PACS | ENL-DEE · PACS, sketch source (ViT) | PG: -0.12 -> 23.59 | Improved |
| use ENL-DEE instead of NTL on DomainNet | ENL-DEE · DomainNet, quickdraw source (ResNet-34) | PG: -38.75 -> 0.49 | Improved |
Closest papers by meaning
- Enhancing Transfer Learning with Flexible Nonparametric Posterior Sampling (ICLR 2024 · similarity 0.87)
- Jointly-Learned Exit and Inference for a Dynamic Neural Network (ICLR 2024 · similarity 0.85)
- Your Transferability Barrier is Fragile: Free-Lunch for Transferring the Non-Transferable Learning (CVPR 2024 · similarity 0.85)
- UniPT: Universal Parallel Tuning for Transfer Learning with Efficient Parameter and Memory (CVPR 2024 · similarity 0.84)
- How to Train Your Multi-Exit Model? Analyzing the Impact of Training Strategies (ICML 2025 · similarity 0.84)
- Dynamic Sparse No Training: Training-Free Fine-tuning for Sparse LLMs (ICLR 2024 · similarity 0.84)
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.
