AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Approximate Equivariance via Projection-Based Regularisation

Torben Berndt, Jan Stühmer

ICML 2026 · Architectures and efficiency › CNN design · approach: Group symmetry equivariant 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
replace sample-based equivariance penalty with projection-based regulariserACDNet · AAPM CT-MARPSNR: 40.02 -> 42.68Improved
replace sample-based penalty with projection-based regulariserACDNet · AAPM CT-MARtraining throughput: 2.54 -> 4.25Improved
add projection-based regulariser to unregularised networkACDNet · AAPM CT-MARPSNR: 42.08 -> 42.68Improved
add projection-based regulariser to unregularised networkOSCNet · AAPM CT-MARPSNR: 42.36 -> 41.88Hurt
add projection-based regulariser to unregularised networkDICDNet · AAPM CT-MARSSIM: 0.9468 -> 0.9605Improved
project linear layers onto equivariant subspace at test time instead of training with the regulariser (ablation)ACDNet · AAPM CT-MARPSNR: 42.68 -> 23.63Hurt
use projection regulariser instead of residual pathway priors (RPP)ACDNet · AAPM CT-MARPSNR: 37.12 -> 42.68Improved
add projection regulariser to relaxed steerable CNNRSteer · Smoke plume rotation DomainMSE: 0.67 -> 0.58Improved
add projection regulariser to relaxed group convRGroup · Smoke plume translation FutureMSE: 0.71 -> 0.72No clear effect
add projection regulariserRGroup · Smoke plume scaling DomainMSE: 0.88 -> 0.88No clear effect
increase non-equivariance penalty weight lambda_perp (ablation)approximately SO(2)-invariant MLP · toy SO(2) classificationempirical equivariance defect: 1.10e+01 -> 1.53e-02Improved
enforce strict SO(3) equivariance instead of partial equivariancesteerable CNN · MedMNIST Organtest accuracy: 0.920 (CNN) -> 0.607Hurt

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