AI Paper Atlas

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

NeurIPS 2020 · Architectures and efficiency › CNN design · approach: Theory of convolutional inductive bias

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 changedWhereMeasuredEffect
Add PAC-Bayes plus weight-correlation term (PBC) as complexity measureComplexity measure ranking · CIFAR-10Kendall's tau: 0.33 -> 0.61Improved
Add PAC-Bayes plus weight-correlation term (PBC) as complexity measureComplexity measure ranking · CIFAR-100Kendall's tau: 0.33 -> 0.55Improved
Use weight correlation alone vs PBC as GE predictorWC measure · CIFAR-10Kendall's tau: 0.61 -> 0.33Hurt
Compare PBC with number of parametersNoP measure · CIFAR-100Kendall's tau: 0.55 -> -0.11Hurt
Add weight correlation descent (WCD) regulariserFCN3 · Fashion-MNISTtest error %: 12.7 -> 12.1Improved
Add WCDVGG16* · Fashion-MNISTtest error %: 13.0 -> 12.4Improved
Add WCDFCN3 · MNISTtest error %: 4.1 -> 3.5Improved
Add WCDVGG16* · MNISTtest error %: 1.4 -> 1.6Hurt
Add WCDVGG19* · MNISTtest error %: 1.6 -> 1.6No clear effect
Add WCDFCN3 · CIFAR-10test error %: 50.7 -> 50.2No clear effect
Add WCDVGG19* · CIFAR-10test error %: 40.5 -> 38.8Improved
Add WCDVGG16* · SVHNtest error %: 20.3 -> 19.6Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy