AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Equivariant Adaptation of Large Pretrained Models

Arnab Kumar Mondal, Siba Smarak Panigrahi, Sékou-Oumar Kaba, Sai Rajeswar, Siamak Ravanbakhsh

NeurIPS 2023 · Probabilistic and scientific modeling › Molecules and proteins · approach: Equivariant and disentangled latent representations · 3 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 prior regularization loss to learned canonicalization (ablation)ResNet50 (LC) · CIFAR10 testaccuracy: 93.29 -> 96.19Improved
add prior regularization loss to learned canonicalization (ablation)ResNet50 (LC) · CIFAR100 testaccuracy: 78.50 -> 83.44Improved
add prior regularization loss to learned canonicalization (ablation)ViT (LC) · CIFAR100 C8-averagedaccuracy: 80.48 -> 83.61Improved
add prior regularization loss to learned canonicalization (ablation)ResNet50 (LC) · STL10 C8-averagedaccuracy: 93.92 -> 96.37Improved
use equivariant canonicalization versus vanilla fine-tuningResNet50 · CIFAR10 C8-averagedaccuracy: 57.77 -> 95.31Improved
use equivariant canonicalization versus vanilla fine-tuningResNet50 · CIFAR10 testaccuracy: 96.97 -> 96.19Hurt
rotation augmentation -180 to +180 versus -10 to +10 for trainingResNet50 pretrained · CIFAR100 testaccuracy: 85.84 -> 80.21Hurt
train from scratch with learned canonicalization instead of small rotation augmentationResNet50 (scratch) · CIFAR10 testaccuracy: 90.96 -> 83.11Hurt
use C8 rotation augmentation instead of full rotation augmentationResNet50 · CIFAR100 C8-averagedaccuracy: 74.12 -> 79.72Improved
make canonicalization network more expressive (G-CNN to G-WRN) (ablation)MaskRCNN + canonicalization · COCO 2017 valmask mAP: 35.77 -> 44.51Improved
make canonicalization network more expressive (G-CNN to G-WRN) (ablation)SAM + canonicalization · COCO 2017 valmask mAP: 59.28 -> 62.13Improved
add prior-regularized canonicalization to MaskRCNNMaskRCNN · COCO 2017 val C4-averagedmask mAP: 27.67 -> 44.50Improved

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