Research map / Architectures and efficiency / CNN design
How does Weight Correlation Affect Generalisation Ability of Deep Neural Networks?
Gaojie Jin, Xinping Yi, Liang Zhang, Lijun Zhang, Sven Schewe, Xiaowei Huang
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 PAC-Bayes plus weight-correlation term (PBC) as complexity measure | Complexity measure ranking · CIFAR-10 | Kendall's tau: 0.33 -> 0.61 | Improved |
| Add PAC-Bayes plus weight-correlation term (PBC) as complexity measure | Complexity measure ranking · CIFAR-100 | Kendall's tau: 0.33 -> 0.55 | Improved |
| Use weight correlation alone vs PBC as GE predictor | WC measure · CIFAR-10 | Kendall's tau: 0.61 -> 0.33 | Hurt |
| Compare PBC with number of parameters | NoP measure · CIFAR-100 | Kendall's tau: 0.55 -> -0.11 | Hurt |
| Add weight correlation descent (WCD) regulariser | FCN3 · Fashion-MNIST | test error %: 12.7 -> 12.1 | Improved |
| Add WCD | VGG16* · Fashion-MNIST | test error %: 13.0 -> 12.4 | Improved |
| Add WCD | FCN3 · MNIST | test error %: 4.1 -> 3.5 | Improved |
| Add WCD | VGG16* · MNIST | test error %: 1.4 -> 1.6 | Hurt |
| Add WCD | VGG19* · MNIST | test error %: 1.6 -> 1.6 | No clear effect |
| Add WCD | FCN3 · CIFAR-10 | test error %: 50.7 -> 50.2 | No clear effect |
| Add WCD | VGG19* · CIFAR-10 | test error %: 40.5 -> 38.8 | Improved |
| Add WCD | VGG16* · SVHN | test error %: 20.3 -> 19.6 | Improved |
Closest papers by meaning
- On the Versatile Uses of Partial Distance Correlation in Deep Learning (ECCV 2022 · similarity 0.86)
- Invariance of Weight Distributions in Rectified MLPs (ICML 2018 · similarity 0.86)
- Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience (ICLR 2019 · similarity 0.86)
- Formalizing Generalization and Adversarial Robustness of Neural Networks to Weight Perturbations (NeurIPS 2021 · similarity 0.86)
- Unveiling The Matthew Effect Across Channels: Assessing Layer Width Sufficiency via Weight Norm Variance (NeurIPS 2024 · similarity 0.85)
- Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study (NeurIPS 2025 · 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.
