AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Learning Layer-wise Equivariances Automatically using Gradients

Tycho F.A. van der Ouderaa, Alexander Immer, Mark van der Wilk

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
train with Laplace marginal likelihood instead of MAPF-FC+CONV · CIFAR-10 testtest accuracy: 81.61 -> 83.92Improved
train with Laplace marginal likelihood instead of MAPFC+CONV · CIFAR-10 testtest accuracy: 76.93 -> 83.32Improved
train with Laplace marginal likelihood instead of MAPS-FC+S-CONV (prior on u) · CIFAR-10 testtest accuracy: 58.80 -> 81.55Improved
train with Laplace marginal likelihood instead of MAPCONV · CIFAR-10 testtest accuracy: 82.81 -> 84.07Improved
train with Laplace marginal likelihood instead of MAPS-CONV (prior on omega) · CIFAR-10 testtest accuracy: 81.84 -> 81.20Hurt
add learnable symmetry (FC path with CONV) versus fixed CONVF-FC+CONV vs CONV · CIFAR-10 testtest accuracy: 84.07 -> 83.92No clear effect
use learnable symmetry model instead of strict CONVF-FC+CONV · Quadrant MNIST digit+quadranttest accuracy: 24.50 -> 99.00Improved
use learnable symmetry model instead of FCF-FC+CONV · Quadrant MNIST digit+quadranttest accuracy: 95.46 -> 99.00Improved
use learnable symmetry instead of strict CONV on strictly symmetric taskF-FC+CONV · Quadrant MNIST digittest accuracy: 99.23 -> 98.86No clear effect
Laplace marginal likelihood for fully connected net instead of MAPFC · CIFAR-10 testtest accuracy: 64.42 -> 52.36Hurt
add rotation-equivariant GCONV path, train with LaplaceF-FC+CONV+GCONV · CIFAR-10test accuracy: 68.33 -> 80.94Improved
add GCONV path, train with LaplaceF-FC+CONV+GCONV · Rotated CIFAR-10test accuracy: 48.30 -> 55.68Improved

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