Research map / Probabilistic and scientific modeling / Variational and generative sampling
On Incorporating Inductive Biases into VAEs
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 |
|---|---|---|---|
| use a clustering-inducing intermediary mapping instead of standard VAE latent construction | InteL-VAE · MNIST | FID: 35.0 ± 0.9 -> 32.2 ±1.5 | Helped |
| use a clustering-inducing intermediary mapping instead of Vamp-VAE | InteL-VAE · MNIST | FID: 38.8 ± 2.4 -> 32.2 ±1.5 | Helped |
| use a clustering-inducing intermediary mapping rather than a mixture-prior baseline | InteL-VAE · MNIST-01 | Uncertainty (%): 4.5 ± 0.8 -> 0.9 ± 0.8 | Helped |
| use a clustering-inducing intermediary mapping rather than Vamp-VAE | InteL-VAE · MNIST-01 | '1' proportion (%): 48.8 ± 0.1 -> 49.5 ± 0.4 | Helped |
| learn a data-dependent sparse mapping with dimension selection | InteL-VAE · CelebA | Sparsity: 0.22±0.01 -> 0.46±0.02 | Helped |
| learn a data-dependent sparse mapping with dimension selection | InteL-VAE (γ=30) · CelebA | FID: 68.6±1.1 -> 64.9±0.4 | Helped |
| increase sparsity regularization weight from γ=30 to γ=70 (ablation) | InteL-VAE · CelebA | Sparsity: 0.25±0.01 -> 0.46±0.02 | Helped |
| increase sparsity regularization weight from γ=30 to γ=70 (ablation) | InteL-VAE · CelebA | FID: 64.9±0.4 -> 68.0±0.6 | Hurt |
| use a hierarchical intermediary mapping | InteL-VAE · CelebA attribute prediction | Accuracy: 0.791 -> 0.817 | Helped |
| use a hierarchical intermediary mapping | InteL-VAE · CelebA attribute prediction | Accuracy: 0.855 -> 0.857 | Helped |
| replace a Gaussian prior directly to induce target structure | VAE · sparse representations | representation sparsity | No clear effect |
| use a conventional sparse-VAE objective rather than a learned dimension-selection mapping | VAE · CelebA | FID: 64.9±0.4 vs 328±10.1; 337±8.1 | Hurt |
Closest papers by meaning
- Demystifying Inductive Biases for (Beta-)VAE Based Architectures (ICML 2021 · similarity 0.92)
- Trading Information between Latents in Hierarchical Variational Autoencoders (ICLR 2023 · similarity 0.90)
- Exploring the Latent Space of Autoencoders with Interventional Assays (NeurIPS 2022 · similarity 0.90)
- Embrace the Gap: VAEs Perform Independent Mechanism Analysis (NeurIPS 2022 · similarity 0.90)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.88)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.88)
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.