Research map / Label-efficient and robust learning / Semi-supervised learning
SHOT-VAE: Semi-supervised Deep Generative Models With Label-aware ELBO 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 |
|---|---|---|---|
| Replace the labeled-data objective with smooth-ELBO (ablation) | Semi-supervised VAE on CIFAR-10 · CIFAR-10 (10% labels) | error rate: 18.08% to 13.54% | Helped |
| Add OT-approximation to smooth-ELBO (ablation) | SHOT-VAE on CIFAR-10 · CIFAR-10 (10% labels) | error rate: 13.54% to 8.51% | Helped |
| Replace the standard labeled-data ELBO with smooth-ELBO (ablation) | Semi-supervised VAE · CIFAR-100 | error rate | Helped |
| Add OT-approximation to the model (ablation) | SHOT-VAE · CIFAR-10 and CIFAR-100 | inference accuracy | Helped |
| Continue optimizing ELBO after its bottleneck without OT-approximation (ablation) | Semi-supervised VAE · CIFAR-10 and CIFAR-100 | inference accuracy | No clear effect |
| Remove the labeled-data ELBO term (ablation) | M2 · CIFAR-10 and CIFAR-100 | inference accuracy | No clear effect |
| Add OT-approximation to smooth-ELBO (ablation) | SHOT-VAE on MNIST · MNIST (100 labels) | error rate: 3.14(\pm 0.19) to 3.12(\pm 0.22) | Helped |
| Add OT-approximation to smooth-ELBO (ablation) | SHOT-VAE on SVHN · SVHN (1k labels) | error rate: 29.38(\pm 0.78) to 28.82(\pm 0.49) | Helped |
| Reduce model size from 36.5M to 1.5M parameters (ablation) | SHOT-VAE · CIFAR-10 (4k labels) | error rate: 6.11(\pm 0.34) to 8.51(\pm 0.32) | Hurt |
| Combine SHOT-VAE with data augmentation (ablation) | SHOT-VAE · CIFAR-10 | accuracy | Helped |
| Combine SHOT-VAE with Mean Teacher (ablation) | SHOT-VAE · CIFAR-100 | accuracy | Helped |
Closest papers by meaning
- VAE Approximation Error: ELBO and Exponential Families (ICLR 2022 · similarity 0.90)
- On the Limitations of Multimodal VAEs (ICLR 2022 · similarity 0.87)
- Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · similarity 0.86)
- Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks (ICML 2017 · similarity 0.86)
- Zero-Shot Learning via Class-Conditioned Deep Generative Models (AAAI 2018 · similarity 0.86)
- PG-LBO: Enhancing High-Dimensional Bayesian Optimization with Pseudo-Label and Gaussian Process Guidance (AAAI 2024 · 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.