AI Paper Atlas

Research map / Learning theory and algorithms / Kernels and regression theory

Scale-invariant Bayesian Neural Networks with Connectivity Tangent Kernel

SungYub Kim, Sihwan Park, Kyung-Su Kim, Eunho Yang

ICLR 2023 · Learning theory and algorithms › Kernels and regression theory · approach: Empirical generalization measures · 0 citations (OpenAlex)

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
Use connectivity-based perturbations to measure sharpnessVGG-13 variants on CIFAR-10 · CIFAR-10 generalization-gap correlationKendall's τ: 0.755 -> 0.837Improved
Use connectivity-based perturbations to measure sharpnessVGG-13 variants on CIFAR-10 · CIFAR-10 generalization-gap correlationGranulated Kendall's Ψ (average): 0.783 -> 0.900Improved
Use connectivity-based perturbations to measure sharpnessVGG-13 variants on CIFAR-10 · CIFAR-10 generalization-gap correlationConditional MI (K): 0.483 -> 0.539Improved
Replace Linearized Laplace with Connectivity LaplaceResNet-18 · CIFAR-100NLL: 1.1673 ± 0.0093 -> 1.1307 ± 0.0042Improved
Replace Linearized Laplace with Connectivity LaplaceResNet-18 · CIFAR-100ECE: 0.0632 ± 0.0010 -> 0.0524 ± 0.0009Improved
Replace Linearized Laplace with Connectivity LaplaceResNet-18 · CIFAR-100Brier score: 0.3597 ± 0.0020 -> 0.3319 ± 0.0005Improved
Replace Linearized Laplace with Connectivity LaplaceResNet-18 · CIFAR-100 OOD detection (SVHN)AUC: 0.8066 ± 0.0120 -> 0.8423 ± 0.0204Improved
Use Connectivity Laplace instead of Linearized LaplaceResNet-18 · CIFAR-100NLL: 1.0110 -> 1.1307 ± 0.0042Hurt
Use Connectivity Laplace instead of Linearized LaplaceResNet-18 · CIFAR-100ECE: 0.0507 -> 0.0524 ± 0.0009Hurt
Use Connectivity Laplace instead of Linearized LaplaceResNet-18 · CIFAR-100Brier score: 0.2740 -> 0.3319 ± 0.0005Hurt
Scale function-preserving parameters preceding batch normalization (ablation)ResNet-18 with PAC-Bayes-CTK · CIFAR-10 and CIFAR-100PAC-Bayes-CTK bound: CIFAR-10: 0.0918 ± 0.0013 -> 0.0907 ± 0.0009; CIFAR-100: 0.2874 ± 0.0034 -> 0.2862 ± 0.0032No clear effect
Use connectivity-based damping rather than parameter-based dampingLaplace approximation with weight decay on batch-normalized networks · Function-preserving weight-decay scalingPredictive uncertainty: LL uncertainty is amplified by 1/γ²; CL uncertainty is preservedImproved

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