Research map / Learning theory and algorithms / Kernels and regression theory
LocalDrop: A Hybrid Regularization for Deep Neural Networks
Ziqing Lu, Chang Xu, Bo Du, Takashi Ishida, Lefei Zhang, Masashi Sugiyama
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 |
|---|---|---|---|
| add LocalDrop | Zhai CNN · CIFAR-10 | error %: 18.05 -> 12.16 | Improved |
| add LocalDrop | Zhai CNN · CIFAR-100 | error %: 50.22 -> 38.10 | Improved |
| apply LocalDrop in convolutional layers (with DropBlock) in addition to FCNs (ablation) | LocalDrop in FCNs · CIFAR-10 | error %: 16.67 -> 12.16 | Improved |
| add dropout to LocalDrop (ablation) | LocalDrop without dropout · CIFAR-10 | error %: 16.12 -> 12.16 | Improved |
| add DropBlock to LocalDrop (ablation) | ResNet-50 LocalDrop without DB · CIFAR-10 | error %: 6.2 -> 5.3 | Improved |
| add DropBlock to LocalDrop (ablation) | ResNet-50 LocalDrop without DB · ILSVRC2012 | error %: 21.9 -> 21.1 | Improved |
| replace DropBlock with LocalDrop | ResNet-50 · ILSVRC2012 | error %: 22.2 -> 21.1 | Improved |
| replace DropBlock with LocalDrop | DPN-92 · CIFAR-10 | error %: 4.5 -> 4.3 | Improved |
| run second-stage low-rank optimization every epoch (m=1) instead of every 10 epochs (ablation) | LocalDrop h=2000 · CIFAR-10 | error %: 12.16 -> 12.80 | Hurt |
| set h=2048 so no singular values are shrunk (ablation) | LocalDrop · CIFAR-10 | error %: 12.16 -> 13.11 | Hurt |
| set lambda to 1 instead of 0.1 (ablation) | LocalDrop h=2000, m=10 · CIFAR-10 | error %: 12.16 -> 12.77 | Hurt |
| use block size 7 instead of 1 (ablation) | LocalDrop ResNet-50 · CIFAR-10 | error %: 6.9 -> 5.3 | Improved |
Closest papers by meaning
- Adaptive Dropout with Rademacher Complexity Regularization (ICLR 2018 · similarity 0.89)
- AutoDropout: Learning Dropout Patterns to Regularize Deep Networks (AAAI 2021 · similarity 0.87)
- On the Regularization Properties of Structured Dropout (CVPR 2020 · similarity 0.86)
- Regularizing Deep Neural Networks by Noise: Its Interpretation and Optimization (NeurIPS 2017 · similarity 0.86)
- Dropout: Explicit Forms and Capacity Control (ICML 2021 · similarity 0.86)
- Dropout Training, Data-dependent Regularization, and Generalization Bounds (ICML 2018 · similarity 0.86)
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.
