Research map / Image generation / Autoregressive and latent generation
Closed-Loop Unsupervised Representation Disentanglement with $\\beta$-VAE Distillation and Diffusion Probabilistic Feedback
Xin Jin, Bohan Li, Baao Xie, Wenyao Zhang, Jinming Liu, Ziqiang Li, Tao Yang, Wenjun Zeng
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 |
|---|---|---|---|
| add beta-VAE distillation and diffusion feedback to Diff-AE | Diff-AE · FFHQ, optical-flow disentanglement metric | changed/frozen ratio (lower is better): 0.1606 -> 0.0177 | Improved |
| add diffusion-feedback dynamic capacity C_dyn on top of knowledge distillation (ablation) | CL-Dis · FFHQ | qualitative disentanglement (Hair direction) | Improved |
| add knowledge distillation from beta-VAE latent (ablation) | Diff-AE with navigation · FFHQ | qualitative disentanglement | Improved |
| add semantic navigation only to Diff-AE latent (ablation) | Diff-AE · FFHQ | qualitative generation quality and disentanglement | Mixed |
| narrow gap between C_max and C_base (ablation) | CL-Dis · Cars3D | FactorVAE score: 0.794 -> 0.762 | Hurt |
| set C_base to 10 instead of 2, 5 or 20 (ablation) | CL-Dis · Cars3D | FactorVAE score: 0.689 (C_base 2) -> 0.794 (C_base 10) | Improved |
| reduce z_sem dimension from 512 to 32 (ablation) | CL-Dis · Shapes3D | FactorVAE score: 0.840 -> 0.837 | No clear effect |
| use CL-Dis semantic encoder instead of learning from scratch for face recognition | face recognition · VGGFace2 | accuracy (%): 81.82 -> 87.32 (5 epochs) | Improved |
| use CL-Dis instead of VAE-based methods for generation quality | CelebA generation · CelebA | FID: 132.80 -> 6.54 | Improved |
| apply CL-Dis to real-recorded MPI3D data | CL-Dis · MPI3D | disentanglement score | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DisDiff — DisDiff: Unsupervised Disentanglement of Diffusion Probabilistic Models (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Factorized Diffusion Autoencoder for Unsupervised Disentangled Representation Learning (AAAI 2024 · similarity 0.90)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.90)
- Can Diffusion Models Disentangle? A Theoretical Perspective (NeurIPS 2025 · similarity 0.89)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.89)
- Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement (NeurIPS 2024 · similarity 0.88)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · 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.
