Research map / Architectures and efficiency / CNN design
Neural Anisotropy Directions
Guillermo Ortiz-Jimenez, Apostolos Modas, Seyed-Mohsen Moosavi, Pascal Frossard
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 |
|---|---|---|---|
| remove pooling layers (ablation) | CNNs on linearly separable Fourier-direction datasets · Linearly separable datasets | test accuracy | Improved |
| increase noise level | CNNs on linearly separable Fourier-direction datasets · Linearly separable datasets | test accuracy | Hurt |
| use a carrier at early NAD indices rather than later indices | ResNet-18 trained on CIFAR-10 with poisoned carriers · CIFAR-10 test set | test accuracy | Mixed |
| flip the CIFAR-10 representation in the NAD basis (ablation) | CNNs trained on CIFAR-10 · CIFAR-10 | test accuracy | Hurt |
| use a random orthonormal basis instead of the 2D-DCT basis to select directions (ablation) | CNNs evaluated on linearly separable datasets · Random-basis linearly separable datasets | test accuracy | No clear effect |
Closest papers by meaning
- Inductive Bias of Deep Convolutional Networks through Pooling Geometry (ICLR 2017 · similarity 0.87)
- Initial Guessing Bias: How Untrained Networks Favor Some Classes (ICML 2024 · similarity 0.86)
- Interpreting Deep Neural Networks with Relative Sectional Propagation by Analyzing Comparative Gradients and Hostile Activations (AAAI 2021 · similarity 0.86)
- The Implicit Bias of Depth: From Neural Collapse to Softmax Codes (ICML 2026 · similarity 0.86)
- Training the Untrainable: Introducing Inductive Bias via Representational Alignment (NeurIPS 2025 · similarity 0.86)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 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.
