Research map / Probabilistic and scientific modeling / Variational and generative sampling
Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks
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 diagonal-Gaussian inference model with a flexible black-box inference model trained adversarially | VAE on the synthetic dataset · Synthetic dataset | log-likelihood: -1.568 -> -1.403 | Helped |
| Replace the diagonal-Gaussian inference model with a flexible black-box inference model trained adversarially | VAE on the synthetic dataset · Synthetic dataset | reconstruction error: 88.5 $\cdot 10^{-3}$ -> 5.77 $\cdot 10^{-3}$ | Helped |
| Add Adaptive Contrast (ablation) | AVB, 8-dimensional latent space · Binarized MNIST | AIS log-likelihood: -91.2 \pm 0.6 -> -89.6 \pm 0.6 | Helped |
| Add Adaptive Contrast (ablation) | AVB, 32-dimensional latent space · Binarized MNIST | AIS log-likelihood: -81.9 \pm 0.4 -> -80.2 \pm 0.4 | Helped |
| Use a 32-dimensional rather than an 8-dimensional latent space with Adaptive Contrast (ablation) | AVB + AC · Binarized MNIST | AIS log-likelihood: -89.6 \pm 0.6 -> -80.2 \pm 0.4 | Helped |
| Use AVB + AC instead of a standard VAE | MNIST model with fully-connected decoder and latent dim 32 · MNIST | AIS: -85.0 \pm 0.3 -> -83.7 \pm 0.3 | Helped |
| Use AVB + AC instead of a standard VAE | MNIST model with convolutional decoder and latent dim 8 · MNIST | AIS: -89.9 \pm 0.4 -> -89.7 \pm 0.3 | No clear effect |
| Use AVB + AC instead of a standard VAE | MNIST model with convolutional decoder and latent dim 32 · MNIST | AIS: -81.9 \pm 0.3 -> -81.7 \pm 0.3 | No clear effect |
| Replace the diagonal-Gaussian inference model with a VAE using inverse autoregressive flow (ablation) | VAE on MNIST with fully-connected decoder · MNIST | approximate ELBO: -88.9 \pm 0.2 -> -88.9 \pm 0.2 | No clear effect |
| Use AVB with Adaptive Contrast instead of the VAE baseline | MNIST model with fully-connected decoder and latent dim 32 · MNIST | reconstruction error: 62.2 \pm 0.2 -> 59.3 \pm 0.2 | Helped |
Closest papers by meaning
- Adversarial Symmetric Variational Autoencoder (NeurIPS 2017 · similarity 0.91)
- On Unifying Deep Generative Models (ICLR 2018 · similarity 0.91)
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.91)
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.91)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.90)
- Trading Information between Latents in Hierarchical Variational Autoencoders (ICLR 2023 · similarity 0.90)
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.