Research map / Architectures and efficiency / RNNs and normalization
Reducing Flipping Errors in Deep Neural Networks
Xiang Deng, Yun Xiao, Bo Long, Zhongfei Zhang
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 |
|---|---|---|---|
| Restrict output-distribution changes on previously correctly classified samples | ResNet-18 · CIFAR-100 | test accuracy: 77.31±0.33 -> 80.29±0.12 | Improved |
| Restrict output-distribution changes on previously correctly classified samples | ResNet-18 · Tiny ImageNet | test accuracy: 65.57±0.16 -> 67.72±0.32 | Improved |
| Restrict output-distribution changes on previously correctly classified samples | ResNet-18 · CUB | test accuracy: 61.09±0.49 -> 67.65±0.25 | Improved |
| Restrict output-distribution changes on previously correctly classified samples | ResNet-18 · Stanford Dogs | test accuracy: 66.56±0.28 -> 70.64±0.25 | Improved |
| Restrict output-distribution changes on previously correctly classified samples | ResNet-18 · CIFAR-100 with 60% label noise | last-epoch test accuracy: 32.22±0.32 -> 61.57±0.08 | Improved |
| Train representations with FER | ResNet-18 · CIFAR-100 to STL-10 | transfer accuracy: 69.71±0.14 -> 70.80±0.15 | Improved |
| Apply behavior regularization to misclassified samples as well (ablation) | FER · CIFAR-100 | test accuracy: 80.29±0.12 -> 80.00±0.18 | Hurt |
| Apply behavior regularization to misclassified samples as well (ablation) | FER · Stanford Dogs | test accuracy: 70.64±0.25 -> 70.25±0.25 | Hurt |
| Use a single correct behavior instead of a weighted average of multiple behaviors (ablation) | FER · CIFAR-100 | test accuracy: 80.29±0.12 -> 79.90±0.07 | Hurt |
| Use a single correct behavior instead of a weighted average of multiple behaviors (ablation) | FER · Stanford Dogs | test accuracy: 70.64±0.25 -> 70.08±0.18 | Hurt |
| Replace FER with LWR, removing both selective regularization and multi-epoch averaging (ablation) | FER · CIFAR-100 | test accuracy: 80.29±0.12 -> 79.73±0.32 | Hurt |
| Use temperature τ = 10 instead of τ = 5 (ablation) | FER with ResNet-18 · CIFAR-100 | test accuracy: 80.29±0.12 -> 79.63±0.15 | Hurt |
Closest papers by meaning
- Shallow-Deep Networks: Understanding and Mitigating Network Overthinking (ICML 2019 · similarity 0.87)
- Computing the Testing Error Without a Testing Set (CVPR 2020 · similarity 0.86)
- Understanding and Utilizing Deep Neural Networks Trained with Noisy Labels (ICML 2019 · similarity 0.86)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.85)
- Repeated Random Sampling for Minimizing the Time-to-Accuracy of Learning (ICLR 2024 · similarity 0.84)
- On Local Overfitting and Forgetting in Deep Neural Networks (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.
