AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

SmoothHess: ReLU Network Feature Interactions via Stein's Lemma

Max Torop, Aria Masoomi, Davin Hill, Kivanc Kose, Stratis Ioannidis, Jennifer Dy

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Width and depth of ReLU networks · 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 SmoothHess + SmoothGrad instead of SoftPlus Hessian + GradientTaylor-expansion explainer · MNIST class logit, eps=0.25P_MSE: 1.2e-6 -> 9.6e-7Improved
Use SmoothHess + SmoothGrad instead of SoftPlus Hessian + GradientTaylor-expansion explainer · CIFAR10 class logit, eps=0.25P_MSE: 2.1e-3 -> 9.8e-4Improved
Use SmoothHess + SmoothGrad instead of SoftPlus Hessian + GradientTaylor-expansion explainer · CIFAR10 penultimate neuron, eps=0.25P_MSE: 1.1e-5 -> 8.1e-7Improved
Add the Hessian term to SmoothGrad (first order to second order) (ablation)Gaussian smoothing · CIFAR10 class logit, eps=0.25P_MSE: 1.3e-2 -> 9.8e-4Improved
Add the Hessian term to SoftPlus gradient (ablation)SoftPlus smoothing · MNIST class logit, eps=1.00P_MSE: 3.9e-4 -> 8.1e-5Improved
Smooth the gradient with SoftPlus vs SmoothGrad (first-order only)first-order explainers · MNIST class logit, eps=0.25P_MSE: 4.6e-6 vs 4.5e-6No clear effect
Use SmoothGrad instead of vanilla gradientfirst-order explainer · MNIST class logit, eps=0.25P_MSE: 4.2e-3 -> 4.5e-6Improved
Choose sigma from only 3 values (eps/sqrt(d) rule) vs SoftPlus beta from over 100 valuesSmoothHess · P_MSE, all datasetsP_MSEImproved
Use SmoothHess + SmoothGrad attack instead of SoftPlus Hessian + Gradient attacksecond-order Taylor attack · MNIST eps=0.75post-hoc accuracy: 51.5 -> 48.0Improved
Use SmoothHess + SmoothGrad attack instead of SoftPlus Hessian + Gradient attacksecond-order Taylor attack · CIFAR10 eps=0.4post-hoc accuracy: 23.5 -> 15.0Improved
Use SmoothHess + SmoothGrad attack instead of vanilla gradient attackTaylor attack · CIFAR10 eps=0.4post-hoc accuracy: 34.0 -> 15.0Improved
Add Hessian term to SmoothGrad for attacks at large radius (ablation)SmoothGrad attack · CIFAR10 eps=1.0post-hoc accuracy: 0.0 -> 4.5Hurt

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