Research map / Architectures and efficiency / RNNs and normalization
MorphNet: Fast & Simple Resource-Constrained Structure Learning of Deep Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply MorphNet structure learning while matching FLOPs | Inception V2 · ImageNet | accuracy: 74.1 -> 75.2 | Helped |
| Apply MorphNet structure learning while matching FLOPs | MobileNet 50% · ImageNet, 128x128 | accuracy: 57.1 -> 58.1 | Helped |
| Apply MorphNet structure learning while matching FLOPs | MobileNet 25% · ImageNet, 128x128 | accuracy: 44.8 -> 45.9 | Helped |
| Apply MorphNet structure learning while matching FLOPs | ResNet101 · JFT | MAP: 0.477 -> 0.487 | Helped |
| Apply MorphNet structure learning while matching FLOPs | AudioResNet · AudioSet | MAP: 0.182 -> 0.186 | Helped |
| Repeat the shrink-expand procedure for a second iteration (ablation) | MorphNet on Inception V2 · ImageNet | accuracy: 74.7% -> 75.2% | Helped |
| Add dropout to the learned architecture to mitigate overfitting (ablation) | MorphNet on Inception V2 · ImageNet | accuracy | Helped |
| Use a width multiplier instead of resource-weighted sparsification for shrinking (ablation) | Inception V2 · ImageNet | accuracy | Hurt |
| Target FLOPs rather than model size with the sparsifying regularizer (ablation) | MorphNet · JFT and AudioSet | MAP: 0.428 and 0.421; baseline model with similar cost is 0.405 | Mixed |
Closest papers by meaning
- Network Morphism (ICML 2016 · similarity 0.88)
- Growing Efficient Deep Networks by Structured Continuous Sparsification (ICLR 2021 · similarity 0.86)
- VanillaNet: the Power of Minimalism in Deep Learning (NeurIPS 2023 · similarity 0.86)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.85)
- Principled Architecture-aware Scaling of Hyperparameters (ICLR 2024 · similarity 0.84)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · 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.