Research map / Optimization / Optimizers
Training Neural Networks Using Features Replay
Zhouyuan Huo, Bin Gu, Heng Huang
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 features replay instead of BP | ResNet164 · CIFAR-10 | testing error (%): 6.40 -> 6.03 | Improved |
| Use features replay instead of BP | ResNet164 · CIFAR-100 | testing error (%): 28.53 -> 27.34 | Improved |
| Use features replay instead of BP | ResNet101 · CIFAR-10 | testing error (%): 5.25 -> 4.97 | Improved |
| Use features replay instead of BP | ResNet101 · CIFAR-100 | testing error (%): 23.48 -> 23.10 | Improved |
| Use features replay instead of BP | ResNet152 · CIFAR-10 | testing error (%): 5.26 -> 4.91 | Improved |
| Use features replay instead of BP | ResNet152 · CIFAR-100 | testing error (%): 25.20 -> 23.61 | Improved |
| Use features replay instead of DDG | ResNet164 · CIFAR-10 | testing error (%): 6.45 -> 6.03 | Improved |
| Use features replay instead of DDG | ResNet164 · CIFAR-100 | testing error (%): 28.51 -> 27.34 | Improved |
| Use features replay instead of DDG | ResNet101 · CIFAR-10 | testing error (%): 5.35 -> 4.97 | Improved |
| Use features replay instead of DDG | ResNet101 · CIFAR-100 | testing error (%): 24.25 -> 23.10 | Improved |
| Use features replay instead of DDG | ResNet152 · CIFAR-10 | testing error (%): 5.72 -> 4.91 | Improved |
| Use features replay instead of DDG | ResNet152 · CIFAR-100 | testing error (%): 26.39 -> 23.61 | Improved |
Closest papers by meaning
- Decoupled Parallel Backpropagation with Convergence Guarantee (ICML 2018 · similarity 0.88)
- Backpropagation-Free Deep Learning with Recursive Local Representation Alignment (AAAI 2023 · similarity 0.86)
- Backprop with Approximate Activations for Memory-efficient Network Training (NeurIPS 2019 · similarity 0.85)
- Accelerated training through iterative gradient propagation along the residual path (ICLR 2025 · similarity 0.85)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (ICML 2024 · similarity 0.84)
- Overcoming Rank Collapse in Feedback Alignment (NeurIPS 2026 · 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.
