AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

LipNeXt: Scaling up Lipschitz-based Certified Robustness to Billion-parameter Models

Kai Hu, Haoqi Hu, Matt Fredrikson

ICLR 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training recipes · 4 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
add periodic polar retraction (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 72.3 -> 73.2Improved
add Lookahead (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 72.4 -> 73.2Improved
add spatial shift on 1/16 of channels (vs none) (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 53.6 -> 73.2Improved
shift 1/16 of channels instead of all channels (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 67.4 -> 73.2Improved
add learnable positional embedding with circular padding (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 72.1 -> 73.2Improved
zero padding instead of circular padding, with positional embedding (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 73.2 -> 72.6Hurt
use beta-Abs (beta=0.75) instead of MinMax (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 72.7 -> 73.2Improved
beta=1.0 instead of 0.75 (ablation)LipNeXt L32W2048 · CIFAR-10CRA at 36/255: 73.2 -> 72.5Hurt
increase width from 1024 to 4096LipNeXt depth 32 · ImageNet-400 classesCRA at eps=1: 22.9 -> 30.0Improved
increase depth from 8 to 64LipNeXt width 2048 · ImageNet-400 classesCRA at eps=1: 22.4 -> 26.9Improved
increase depth from 64 to 128LipNeXt width 2048 · ImageNet-400 classesCRA at eps=1: 26.9 -> 26.8No clear effect
add diffusion-generated synthetic data and scale to L32W2896LipNeXt · CIFAR-10CRA at 36/255: 73.2 -> 81.7Improved

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