Research map / Probabilistic and scientific modeling / Variational and generative sampling
AVT: Unsupervised Learning of Transformation Equivariant Representations by Autoencoding Variational Transformations
Guo-Jun Qi, Liheng Zhang, Chang Wen Chen, Qi Tian
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 |
|---|---|---|---|
| replace AET transformation prediction with AVT variational mutual-information objective | NIN unsupervised representation + conv classifier · CIFAR-10 | error rate (%): 7.82 -> 7.75 | Improved |
| use AVT instead of AET-project with non-linear classifier | NIN · CIFAR-10 | error rate (%): 9.41 -> 8.96 | Improved |
| use AVT instead of AET-project with 1 FC classifier | NIN · CIFAR-10 (1 FC) | error rate (%): 16.65 -> 16.19 | Improved |
| use AVT instead of AET-project with 3 FC classifier | NIN · CIFAR-10 (3 FC) | error rate (%): 9.92 -> 9.55 | Improved |
| use AVT instead of AET-project for KNN with K=5 | NIN features · CIFAR-10 KNN | error rate (%): 22.40 -> 21.62 | Improved |
| use AVT instead of AET-project for KNN with K=10 | NIN features · CIFAR-10 KNN | error rate (%): 22.39 -> 23.7 | Hurt |
| use AVT instead of AET-project for KNN with K=20 | NIN features · CIFAR-10 KNN | error rate (%): 23.73 -> 21.51 | Improved |
| use AVT unsupervised pretraining instead of fully supervised training with 1000 labels per class | NIN conv classifier · CIFAR-10 (limited labels) | error rate (%): 16.53 -> 12.27 | Improved |
| use AVT unsupervised pretraining instead of AET-project with 20 labels per class | NIN conv classifier · CIFAR-10 (20 labels per class) | error rate (%): 34.83 -> 35.44 | Hurt |
| use the 13-layer network instead of NIN | AVT + conv · CIFAR-10 (20 labels per class) | error rate (%): 35.44 -> 26.2 | Improved |
| use AVT instead of AET-project, non-linear classifier on Conv4 | AlexNet · ImageNet | top-1 accuracy (%): 53.2 -> 54.2 | Improved |
| use AVT instead of AET-project, non-linear classifier on Conv5 | AlexNet · ImageNet | top-1 accuracy (%): 47.0 -> 48.4 | Improved |
Closest papers by meaning
- Learning Generalized Transformation Equivariant Representations Via AutoEncoding Transformations (TPAMI 2020 · similarity 0.96)
- AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations Rather Than Data (CVPR 2019 · similarity 0.90)
- Unsupervised Variational Translator for Bridging Image Restoration and High-Level Vision Tasks (ECCV 2024 · similarity 0.88)
- Self-supervised Transformation Learning for Equivariant Representations (NeurIPS 2024 · similarity 0.88)
- Non-Parametric Transformation Networks for Learning General Invariances from Data (AAAI 2019 · similarity 0.88)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · similarity 0.86)
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.
