Research map / Probabilistic and scientific modeling / Variational and generative sampling
Adversarial Disentanglement with Grouped Observations
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 |
|---|---|---|---|
| Add adversarial mutual-information minimization to MLVAE | MLVAE · MNIST test set, K=2, dc=2, ds=14 | content classification accuracy: 65.0% -> 97.6% | Helped |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · MNIST test set, K=2, dc=8, ds=8 | style classification accuracy: 79.6% -> 26.8% | Helped |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · Chairs test set, K=2 | content classification accuracy: 76.3% -> 93.8% | Helped |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · Chairs test set, K=2 | reconstruction error: 85.7 -> 111.7 | Hurt |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · VGGFace2 test set, K=2 | content classification accuracy: 21.4% -> 39.0% | Helped |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · VGGFace2 test set, K=2 | style classification accuracy: 27.5% -> 6.8% | Helped |
| Add adversarial mutual-information minimization to MLVAE | MLVAE · MNIST-ROT, ±65° | content classification accuracy: 29.3% -> 71.0% | Helped |
| Increase style regularization weight β from 1.5 to 5.0 (ablation) | MLVAE · MNIST test set, K=2, dc=2, ds=14 | content classification accuracy: 87.6% -> 95.6% | Helped |
| Increase style regularization weight β from 1.5 to 5.0 (ablation) | MLVAE · MNIST test set, K=2, dc=2, ds=14 | reconstruction error: 80.4 -> 106.8 | Hurt |
| Increase style regularization weight β from 5.0 to 20.0 (ablation) | MLVAE · MNIST test set, K=2, dc=2, ds=14 | content classification accuracy: 95.6% -> 94.5% | Mixed |
| Increase style regularization weight β from 5.0 to 20.0 (ablation) | MLVAE · MNIST test set, K=2, dc=2, ds=14 | reconstruction error: 106.8 -> 141.3 | Hurt |
Closest papers by meaning
- Simple Disentanglement of Style and Content in Visual Representations (ICML 2023 · similarity 0.89)
- Multi-Level Variational Autoencoder: Learning Disentangled Representations From Grouped Observations (AAAI 2018 · similarity 0.88)
- Unsupervised Disentanglement of Content and Style via Variance-Invariance Constraints (ICLR 2025 · similarity 0.87)
- Demystifying Inter-Class Disentanglement (ICLR 2020 · similarity 0.87)
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.87)
- Factorized Diffusion Autoencoder for Unsupervised Disentangled Representation Learning (AAAI 2024 · similarity 0.85)
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.