Research map / Architectures and efficiency / CNN design
Building a Regular Decision Boundary With Deep 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 |
|---|---|---|---|
| Increase network width from K=16 to K=512 (ablation) | 13-layer convolutional network · CIFAR10 | accuracy: 79.8% -> 95.4% | Helped |
| Increase network width from K=16 to K=512 (ablation) | 13-layer convolutional network · CIFAR100 | accuracy: 30.6% -> 79.6% | Helped |
| Replace ReLU with sign(x)(sqrt(|x|)+0.1) (ablation) | CNN with K=256 · CIFAR10 | accuracy: 89.0% | No clear effect |
| Apply ReLU to all feature maps instead of varying the nonlinear fraction (ablation) | CNN with K=32 · CIFAR10 | accuracy: 88.0% versus maximum 89.8% | No clear effect |
| Apply ReLU to all feature maps instead of varying the nonlinear fraction (ablation) | CNN with K=128 · CIFAR10 | accuracy: 94.7% versus maximum 94.7% | No clear effect |
| Make the network almost linear by setting k=0 (ablation) | ReLU^K_k CNN · CIFAR10 | accuracy: roughly 70% | Hurt |
| Increase the fraction of feature maps receiving ReLU to at least 0.6 (ablation) | ReLU^K_k CNN · CIFAR10 | accuracy | No clear effect |
| Use wider layers (ablation) | 13-layer convolutional network · CIFAR10 and CIFAR100 | accuracy | Helped |
Closest papers by meaning
- Understanding deep learning requires rethinking generalization (ICLR 2017 · similarity 0.88)
- Hold me tight! Influence of discriminative features on deep network boundaries (NeurIPS 2020 · similarity 0.87)
- Empirical Study of the Topology and Geometry of Deep Networks (CVPR 2018 · similarity 0.87)
- Tensor Programs VI: Feature Learning in Infinite Depth Neural Networks (ICLR 2024 · similarity 0.86)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.86)
- Mind the Boundary: Coreset Selection via Reconstructing the Decision Boundary (ICML 2024 · 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.