Research map / Architectures and efficiency / RNNs and normalization
Depth-Progressive Monotonic Learning without Global Backpropagation
Chenhao Ye, Rongguang Ye, Yuchao Zhang, Ming Tang
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 |
|---|---|---|---|
| replace global backprop with DMoL local belief objectives | SimpleCNN · CIFAR-100 | accuracy (%): 51.81 -> 56.12 | Improved |
| replace global backprop with DMoL | SimpleCNN · Tiny-ImageNet | accuracy (%): 34.51 -> 41.37 | Improved |
| replace global backprop with DMoL | SimpleCNN / ViT-Tiny · CIFAR-10 and ImageNet | accuracy (%): 84.32 -> 84.42 (CIFAR-10); 63.68 -> 63.74 (ImageNet) | No clear effect |
| replace BP with DMoL in 8-layer GCN to reduce over-smoothing | 8-layer GCN · Cora | accuracy (%): 43.86 -> 81.40 | Improved |
| replace BP with DMoL in 8-layer GCN | 8-layer GCN · Actor | accuracy (%): 25.46 -> 27.78 | Improved |
| replace BP with DMoL for diffusion transformer training | DiT · CIFAR-100 label-to-image | final training MSE: 0.060 -> 0.034 | Improved |
| use standard ResNet-18 with DMoL instead of BP | ResNet-18 · CIFAR-10 | accuracy (%): 95.68 -> 93.32 | Hurt |
| adapt ResNet-18 to DMoL (ResNet-DMoL-RA variant) (ablation) | DMoL on ResNet-18 · CIFAR-100 | accuracy (%): 73.02 -> 79.16 | Improved |
| increase network depth from 4 to 20 modules (ablation) | DMoL vs BP · CIFAR-100 | accuracy (%) | Mixed |
| remove then restore one-third of intermediate modules during training (ablation) | BP vs DMoL · structural fragility test | test accuracy | Mixed |
| add parallel workers for module updates | DMoL vs BP · CIFAR-100 | throughput speedup | Improved |
Closest papers by meaning
- Adaptive Depth Networks with Skippable Sub-Paths (NeurIPS 2024 · similarity 0.84)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.84)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.84)
- Dendritic Localized Learning: Toward Biologically Plausible Algorithm (ICML 2025 · similarity 0.84)
- Dynamic Sparse No Training: Training-Free Fine-tuning for Sparse LLMs (ICLR 2024 · similarity 0.84)
- $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks (NeurIPS 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.
