AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Equivariant Neural Networks for General Linear Symmetries on Lie Algebras

Chankyo Kim, Sicheng Zhao, Minghan Zhu, Tzu-Yuan Lin, Maani Ghaffari

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Symmetry and equivariance in loss landscapes

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 the semisimple ideal to a center-only model (full reductive structure) (ablation)ReLN (v, log C) · Drone state estimationATE (m): 16.27 -> 13.92Improved
add the central component to a semisimple-only (Lie Neurons style) model (ablation)ReLN · Drone state estimationATE (m): 16.86 -> 16.49Improved
use log-covariance instead of raw covariance as inputReLN · Drone state estimationATE (m): 16.49 -> 13.92Improved
add covariance C to velocity inputReLN · Drone state estimationATE (m): 16.85 -> 16.49Improved
add covariance C to velocity input via eigendecompositionVN · Drone state estimationATE (m): 17.36 -> 191.78Hurt
replace non-equivariant ResNet with ReLNResNet · Drone state estimationATE (m): 205.11 -> 16.49Improved
replace MLP with ReLN under camera rotationMLP · Platonic solid classification (rotated camera)accuracy (%): 36.54% -> 99.78%Improved
add SO(3) augmentation to MLPMLP · Platonic solid classification (in-distribution)accuracy (%): 95.76% -> 81.47%Hurt
double MLP widthMLP · Platonic solid classification (rotated camera)accuracy (%): 36.54% -> 36.55%No clear effect
replace Lie Neurons with ReLNLie Neurons · sp(4) invariant regressiontest MSE: 5.83x10^-4 vs 5.14x10^-4No clear effect
add Sp(4) conjugation augmentation to MLP 512MLP 512 · sp(4) invariant regression, Sp(4)-conjugated testtest MSE: 0.906 -> 0.446Improved
replace baseline Gaussian-MAE with ReLN encoder/decoderGaussian-MAE · ModelNet10 classification, randomly rotated testaccuracy (%): 18.28 -> 95.15Improved

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