AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Learning Probabilistic Symmetrization for Architecture Agnostic Equivariance

Jinwoo Kim, Dat Tien Nguyen, Ayhan Suleymanzade, Hyeokjun An, Seunghoon Hong

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Symmetry and equivariance in loss landscapes · 2 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 uniform group averaging with a learned probabilistic distributionMLP symmetrization · EXP-classifytest accuracy: 50% -> 100%Improved
use frame averaging instead of learned probabilistic symmetrizationMLP symmetrization · EXP-classifytest accuracy: 100% -> 100%No clear effect
add stochastic sampling rather than fixing ε at initialization (ablation)MLP-PS · EXP-classifytest accuracy: 79.5% -> 100%Improved
add probabilistic sampling instead of using a deterministic canonicalizer (ablation)MLP symmetrization · EXP-classifytest accuracy: 50% -> 100%Improved
learn the symmetrization distribution instead of using uniform group averagingTransformer symmetrization · n-bodyPosition MSE: 0.00414 ± 0.00001 -> 0.00401 ± 0.00001Improved
add probabilistic sampling instead of one deterministic group element per input (ablation)Transformer symmetrization · n-bodyPosition MSE: 0.00508 -> 0.00401 ± 0.00001Improved
add 10-sample symmetrization at test time instead of 1 sample (ablation)ViT-PS without pretraining · PATTERNtest accuracy: 85.868 ± 0.017 -> 85.989 ± 0.011Improved
add ImageNet-21k pretraining (ablation)ViT-PS, 10-sample · PATTERNtest accuracy: 85.989 ± 0.011 -> 86.650 ± 0.010Improved
add 10-sample symmetrization at test time instead of 1 sample (ablation)ViT-PS pretrained on ImageNet-21k · PATTERNtest accuracy: 86.573 ± 0.030 -> 86.650 ± 0.010Improved
use learned probabilistic symmetrization instead of canonicalizationViT pretrained on ImageNet-21k · PATTERNtest accuracy: 86.534 -> 86.650 ± 0.010Improved
use ImageNet-21k pretrained ViT with probabilistic symmetrization rather than the strongest listed graph baselinePATTERN node classification · PATTERNtest accuracy: 86.245 -> 86.650 ± 0.010Improved
use ViT-PS instead of GraphGPSPeptides-func · Peptides-funcAP: 0.6535 -> 0.6575Improved

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