Research map / Architectures and efficiency / RNNs and normalization
Deep Elastic Networks With Model Selection for Multi-Task Learning
Chanho Ahn, Eunwoo Kim, Songhwai Oh
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 |
|---|---|---|---|
| use DEN with rho=0.1 instead of separate baseline models on three tasks | ResNet-42 · STL-10 (M1) | accuracy (%): 76.25 -> 87.58 | Improved |
| use DEN (rho=0.1) instead of baseline | ResNet-42 · Tiny-ImageNet (M1) | accuracy (%): 57.22 -> 60.21 | Improved |
| use DEN (rho=0.1) with one shared network instead of three baselines | ResNet-42 · M1 average of three tasks | accuracy (%): 69.51 -> 74.30 | Improved |
| use DEN (rho=0.1) instead of baseline | WRN-32-4 · M1 average of three tasks | accuracy (%): 71.26 -> 73.82 | Improved |
| raise sparsity weight rho from 0.1 to 1 (ablation) | DEN · M1 average (ResNet-42) | accuracy (%): 74.30 -> 71.65 | Hurt |
| raise rho from 0.1 to 1 (ablation) | DEN · M1 FLOPs (ResNet-42) | FLOPs: 1.61G -> 1.35G | Improved |
| use DEN (rho=0.1) instead of Routing | Mini-ImageNet MTL · Mini-ImageNet (M2, 224x224) | accuracy (%): 58.97 -> 65.23 | Improved |
| use DEN (rho=1) instead of baseline on coarse labels | WRN-32-4 · CIFAR-100 coarse (M3) | accuracy (%): 83.53 -> 84.87 | Improved |
| use DEN (rho=1) instead of baseline on fine labels | WRN-32-4 · CIFAR-100 fine (M3) | accuracy (%): 76.32 -> 75.93 | No clear effect |
| add the learned selector instead of a random selector (ablation) | DEN ResNeXt-29 (8x64d) · CIFAR-10 compression (C1) | accuracy (%): 90.55 -> 94.61 | Improved |
| add the learned selector instead of a random selector (ablation) | DEN ResNeXt-29 (4x64d) · CIFAR-100 compression (C1) | accuracy (%): 65.44 -> 77.58 | Improved |
| increase levels of hierarchy per block up to four, then to five (ablation) | DEN WRN-32-4 · CIFAR-100 | accuracy | Mixed |
Closest papers by meaning
- ELASTIC: Improving CNNs With Dynamic Scaling Policies (CVPR 2019 · similarity 0.87)
- Blockout: Dynamic Model Selection for Hierarchical Deep Networks (CVPR 2016 · similarity 0.85)
- Hypernetworks for Dynamic Feature Selection (NeurIPS 2026 · similarity 0.85)
- Elastic ViTs from Pretrained Models without Retraining (NeurIPS 2025 · similarity 0.84)
- Cluster-Learngene: Inheriting Adaptive Clusters for Vision Transformers (NeurIPS 2024 · similarity 0.83)
- When Dynamic Data Selection Meets Data Augmentation: Achieving Enhanced Training Acceleration (ICML 2025 · similarity 0.83)
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.
