Research map / Architectures and efficiency / RNNs and normalization
Feedback Networks
Check 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 |
|---|---|---|---|
| add temporal skip connections (ablation) | Feedback Net · CIFAR-100 | Top1 accuracy: 67.37 -> 67.83 | Helped |
| add temporal skip connections (ablation) | Feedback Net · CIFAR-100 | Top5 accuracy: 89.97 -> 90.12 | Helped |
| use feedback module length 2 rather than length 1 (ablation) | Feedback Net · CIFAR-100 | Top1 accuracy: 66.29 -> 67.83 | Helped |
| use feedback module length 2 rather than Stack-All (ablation) | Feedback Net · CIFAR-100 | Top1 accuracy: 65.85 -> 67.83 | Helped |
| train with episodic curriculum learning (ablation) | Feedback Net · CIFAR-100 | Top1 accuracy: 68.21 -> 69.57 | Helped |
| train with episodic curriculum learning (ablation) | Feedforward ResNet with auxiliary loss · CIFAR-100 | Top1 accuracy: 69.36 -> 69.24 | Hurt |
| train with conventional curriculum learning (ablation) | Feedforward ResNet without auxiliary loss · CIFAR-100 | Top1 accuracy: 69.36 -> 65.69 | Hurt |
| train with curriculum learning (ablation) | Feedback Net · Stanford Cars | Fine accuracy: 50.33 -> 53.37 | Helped |
| train with curriculum learning (ablation) | Feedback Net · Stanford Cars | Coarse accuracy: 74.15 -> 80.7 | Helped |
| replace feedforward convolutional layers with feedback iterations | Hourglass · MPII Human Pose | PCKh: 77.6 -> 82.3 | Helped |
| use feedback instead of a parallel ResNet ensemble for early predictions | CIFAR-100 classifier · CIFAR-100 at 12T | Top1 accuracy: 66.35 -> 67.94 | Helped |
| use feedback instead of a parallel ResNet ensemble for early predictions | CIFAR-100 classifier · CIFAR-100 at 21T | Top1 accuracy: 68.2 -> 71.12 | Helped |
Closest papers by meaning
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.86)
- Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization (ICLR 2024 · similarity 0.86)
- Curriculum Learning by Transfer Learning: Theory and Experiments with Deep Networks (ICML 2018 · similarity 0.85)
- Direct Feedback Alignment Provides Learning in Deep Neural Networks (NeurIPS 2016 · similarity 0.85)
- Feedback-Prop: Convolutional Neural Network Inference Under Partial Evidence (CVPR 2018 · similarity 0.85)
- Feedback Network for Image Super-Resolution (CVPR 2019 · similarity 0.85)
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.