Research map / Probabilistic and scientific modeling / Variational and generative sampling
Data Augmentation in High Dimensional Low Sample Size Setting Using a Geometry-Based Variational Autoencoder
Clément Chadebec, Elina Thibeau–Sutre, Ninon Burgos, Stéphanie Allassonnière
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Hamiltonian flows and Riemannian metric to VAE | VAE · MNIST test | log-likelihood: -92.94 -> -82.64 | Improved |
| add geometry-aware metric to HVAE | HVAE · MNIST test | log-likelihood: -85.33 -> -82.64 | Improved |
| add Riemannian metric to HVAE | HVAE · MNIST test | KL to true posterior: 3.61 -> 3.57 | Improved |
| sample with metric-volume distribution instead of prior | RHVAE · reduced MNIST (balanced) | GAN-train: 82.0 -> 90.1 | Improved |
| sample with metric-volume distribution instead of prior | RHVAE · reduced MNIST (balanced) | GAN-test: 63.1 -> 88.1 | Improved |
| sample with metric-volume distribution instead of prior | RHVAE · reduced MNIST (unbalanced) | GAN-train: 69.3 -> 86.2 | Improved |
| replace prior sampling with GMM(20) on RAE vs proposed | RAE · reduced EMNIST | GAN-train: 82.6 -> 79.5 | Hurt |
| increase GMM components in RAE | RAE · reduced MNIST (balanced) | GAN-test: 91.7 -> 96.6 | Hurt |
| add 1k synthetic samples per class to real data | Ours · CIFAR (reduced) | accuracy: 42.6 -> 49.2 | Improved |
| add 1k synthetic samples per class to real data | Ours · MNIST unbalanced (reduced) | accuracy: 81.5 -> 89.7 | Improved |
| train only on 1k synthetic samples per class | Ours · MNIST (reduced) | accuracy: 89.9 -> 90.1 | Improved |
| add 1k VAE-GMM samples to real data | VAE GMM · CIFAR (reduced) | accuracy: 42.6 -> 38.9 | Hurt |
Closest papers by meaning
- Learning Weighted Submanifolds With Variational Autoencoders and Riemannian Variational Autoencoders (CVPR 2020 · similarity 0.87)
- Sparse Multi-Channel Variational Autoencoder for the Joint Analysis of Heterogeneous Data (ICML 2019 · similarity 0.86)
- Graph Geometry-Preserving Autoencoders (ICML 2024 · similarity 0.86)
- A Statistical Analysis of Wasserstein Autoencoders for Intrinsically Low-dimensional Data (ICLR 2024 · similarity 0.85)
- Sparse Autoencoders, Again? (ICML 2025 · similarity 0.84)
- Curvature Enhanced Data Augmentation for Regression (ICML 2025 · similarity 0.84)
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.
