Research map / Probabilistic and scientific modeling / Variational and generative sampling
ControlVAE: Tuning, Analytical Properties, and Performance Analysis
Huajie Shao, Zhisheng Xiao, Shuochao Yao, Dachun Sun, Aston Zhang, Shengzhong Liu, Tianshi Wang, Jinyang Li, Tarek Abdelzaher
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 |
|---|---|---|---|
| use PI-controlled KL weight instead of fixed-weight VAE | VAE · CIFAR-10 | ELBO: -372.10 -> -365.18 | Improved |
| use PI-controlled KL weight instead of fixed-weight VAE | VAE · CIFAR-10 | FID: 135.25 -> 122.36 | Improved |
| use PI controller instead of Lagrange multiplier | ControlVAE · CIFAR-10 | ELBO: -366.87 -> -365.18 | Improved |
| use PI controller instead of fixed large beta | beta-VAE · CIFAR-10 | ELBO: -388.82 -> -365.18 | Improved |
| use ControlVAE instead of VAE | VAE · CelebA | FID: 89.33 -> 86.91 | Improved |
| use ControlVAE instead of LM | Lagrange multiplier · CelebA | FID: 87.01 -> 86.91 | No clear effect |
| raise target KL from 25 to 35 in dialog generation | ControlVAE · Switchboard | Dis-2: 83.15K -> 95.86K | Improved |
| use ControlVAE-KL-35 instead of cost annealing | Cost annealing · Switchboard | PPL: 16.82 -> 8.81 | Improved |
| raise target KL from 16 to 19 (ablation) | ControlVAE · 2D Shapes | RMIG: 0.4637 -> 0.4218 | Hurt |
| set annealing step s=0.15 instead of 0.2 (ablation) | ControlVAE · 2D Shapes | RMIG: 0.4304 -> 0.4637 | Improved |
| use batch size 100 instead of 50 (ablation) | ControlVAE · CIFAR-10 | ELBO: -375.99 -> -368.18 | Improved |
| use embedding size 200 instead of 100 (ablation) | ControlVAE · CIFAR-10 | ELBO: -365.18 -> -361.80 | Improved |
Closest papers by meaning
- Rethinking Controllable Variational Autoencoders (CVPR 2022 · similarity 0.95)
- ControlVAE: Controllable Variational Autoencoder (ICML 2020 · similarity 0.95)
- Variational Autoencoder with Implicit Optimal Priors (AAAI 2019 · similarity 0.89)
- Soft-IntroVAE: Analyzing and Improving the Introspective Variational Autoencoder (CVPR 2021 · similarity 0.89)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.89)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.89)
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.
