Research map / Architectures and efficiency / RNNs and normalization
Training Neural Networks with Local Error Signals
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 |
|---|---|---|---|
| combine local prediction and similarity-matching losses (ablation) | VGG11B on CIFAR-10 with standard data augmentation · CIFAR-10 | test error (%): 6.70 -> 5.30 | Helped |
| combine local prediction and similarity-matching losses (ablation) | VGG8B on Fashion-MNIST · Fashion-MNIST | test error (%): 5.66 -> 4.65 | Helped |
| add cutout to VGG8B trained with predsim (ablation) | VGG8B on STL-10 · STL-10 | test error (%): 20.51 -> 19.25 | Helped |
| add cutout to VGG8B(2x) trained with predsim (ablation) | VGG8B(2x) on Fashion-MNIST · Fashion-MNIST | test error (%): 4.33 -> 4.14 | Helped |
| add cutout to VGG8B trained with predsim (ablation) | VGG8B on SVHN · SVHN | test error (%): 1.74 -> 1.65 | Helped |
| add cutout to VGG11B(3x) trained with predsim (ablation) | VGG11B(3x) on CIFAR-10 · CIFAR-10 | test error (%): 3.97 -> 3.60 | Helped |
| add cutout to VGG8B trained with predsim (ablation) | VGG8B on Kuzushiji-MNIST · Kuzushiji-MNIST | test error (%): 1.36 -> 0.99 | Helped |
| add cutout to the proposed method (ablation) | CIFAR-100 models · CIFAR-100 | test error (%) | No clear effect |
| combine feedback-alignment prediction loss with backprop-free similarity matching (ablation) | VGG8B on CIFAR-10 without back-propagation · CIFAR-10 | test error (%): 9.80 -> 9.02 | Helped |
| add local similarity matching to global back-propagation (ablation) | VGG8B on SVHN · SVHN | test error (%): 2.29 -> 1.95 | Hurt |
| add local similarity matching to global back-propagation (ablation) | VGG8B on STL-10 · STL-10 | test error (%): 33.1 -> 25.6 | Hurt |
| replace max-pooling layers with stride-2 convolutional layers (ablation) | VGG-like local-loss networks · image classification experiments | test error | Hurt |
Closest papers by meaning
- Towards Interpretable Deep Local Learning with Successive Gradient Reconciliation (ICML 2024 · similarity 0.87)
- Understanding the Loss Surface of Neural Networks for Binary Classification (ICML 2018 · similarity 0.86)
- Local Loss Optimization in the Infinite Width: Stable Parameterization of Predictive Coding Networks and Target Propagation (ICLR 2025 · similarity 0.85)
- Global Optimality in Neural Network Training (CVPR 2017 · similarity 0.85)
- Dendritic Localized Learning: Toward Biologically Plausible Algorithm (ICML 2025 · similarity 0.85)
- Scaling Supervised Local Learning with Augmented Auxiliary Networks (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.