Research map / Probabilistic and scientific modeling / Variational and generative sampling
Demystifying Inductive Biases for (Beta-)VAE Based Architectures
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 original images with locally manipulated images | β-VAE · dSprites | MIG: 0.23 \pm 0.08 -> 0.07\pm0.09 | Hurt |
| replace the original images with locally manipulated images | β-VAE · Shapes3D | MIG: 0.60 \pm 0.31 -> 0.09 \pm 0.14 | Hurt |
| replace the original images with locally manipulated images | FactorVAE · dSprites and Shapes3D | MIG: dSprites: 0.27 \pm 0.11 -> 0.20 \pm 0.12; Shapes3D: 0.27 \pm 0.18 -> 0.07 \pm 0.05 | Mixed |
| replace the original images with locally manipulated images | TC-β-VAE · dSprites and Shapes3D | MIG: dSprites: 0.25 \pm 0.08 -> 0.14\pm0.10; Shapes3D: 0.58 \pm 0.20 -> 0.24 \pm 0.16 | Hurt |
| replace the original images with locally manipulated images | Slow-VAE · dSprites and Shapes3D | MIG: dSprites: 0.39 \pm 0.08 -> 0.27 \pm 0.08; Shapes3D: 0.53 \pm 0.19 -> 0.13 \pm 0.08 | Hurt |
| replace the original images with locally manipulated images | PCL · dSprites and Shapes3D | MIG: dSprites: 0.21 \pm 0.03 -> 0.24\pm 0.07; Shapes3D: 0.44 \pm 0.06 -> 0.47 \pm 0.08 | No clear effect |
| replace structured manipulation with uniform pixel-wise noise | β-VAE · Shapes3D | MIG: 0.60 \pm 0.31 -> 0.66 \pm 0.05 | No clear effect |
| replace structured manipulation with uniform pixel-wise noise | β-VAE · dSprites | MIG: 0.23 \pm 0.08 -> 0.14 \pm 0.07 | Hurt |
| retune the primary regularization hyperparameter | VAE-based architectures on Shapes3D modified · Shapes3D modified | MIG | Mixed |
Closest papers by meaning
- On Incorporating Inductive Biases into VAEs (ICLR 2022 · similarity 0.92)
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.92)
- An Identifiable Double VAE For Disentangled Representations (ICML 2021 · similarity 0.92)
- beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework (ICLR 2017 · similarity 0.91)
- Tripod: Three Complementary Inductive Biases for Disentangled Representation Learning (ICML 2024 · similarity 0.90)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · 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.