Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Invariance Learning in Deep Neural Networks with Differentiable Laplace Approximations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| learn affine invariances with LILA instead of Augerino | ResNet · CIFAR-10 fully rotated | test accuracy: 75.40 ± 0.19 -> 79.50 ± 0.62 | Helped |
| learn affine invariances with LILA instead of Augerino | ResNet · CIFAR-10 translated | test accuracy: 73.71 ± 0.31 -> 79.21 ± 0.17 | Helped |
| learn affine invariances with LILA instead of Augerino | ResNet · CIFAR-10 original | test accuracy: 79.03 ± 1.04 -> 84.19 ± 0.76 | Helped |
| learn affine invariances with LILA instead of Augerino | MLP · FashionMNIST fully rotated | test accuracy: 77.76 ± 0.15 -> 87.39 ± 0.06 | Helped |
| learn affine invariances with LILA instead of Augerino | MLP · MNIST partially rotated | test accuracy: 96.35 ± 0.02 -> 97.81 ± 0.11 | Helped |
| learn invariances with LILA rather than use a non-invariant model | Wide ResNet · CIFAR-10 | test accuracy: 85.17 ± 0.39 -> 91.98 ± 0.04 | Helped |
| replace GGN with empirical Fisher (ablation) | LILA, Wide ResNet · CIFAR-10 | test accuracy: 91.98 ± 0.04 -> 91.64 ± 0.26 | Hurt |
| learn invariances on 1000 training examples rather than use a non-invariant model | LILA · CIFAR-10 subset of 1000 samples | test accuracy: +22 percentage points | Helped |
| use LILA with 10% of the training data rather than the non-invariant model's full data | LILA · CIFAR-10 fully rotated | data efficiency: 10% | Helped |
| use diagonal GGN or empirical Fisher instead of KFAC-GGN (ablation) | LILA · image classification invariance learning | test performance | Hurt |
Closest papers by meaning
- Deep invariant networks with differentiable augmentation layers (NeurIPS 2022 · similarity 0.90)
- Learning Invariances using the Marginal Likelihood (NeurIPS 2018 · similarity 0.90)
- Learning Invariances in Neural Networks from Training Data (NeurIPS 2020 · similarity 0.89)
- Automatic Data Augmentation via Invariance-Constrained Learning (ICML 2023 · similarity 0.87)
- A Group-Theoretic Framework for Data Augmentation (NeurIPS 2020 · similarity 0.87)
- First-Order Manifold Data Augmentation for Regression Learning (ICML 2024 · similarity 0.85)
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.