Research map / Label-efficient and robust learning / Recognition and long-tail
PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS
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 |
|---|---|---|---|
| Progressively grow hierarchical representations from high to low abstraction levels | VLAE · 3DShapes and dSprites | MIG and MIG-sup | Helped |
| Progressively grow hierarchical representations from high to low abstraction levels | VLAE · 3DShapes and dSprites | factor metric | Mixed |
| Use a smaller beta (beta < 20) (ablation) | pro-VLAE · Disentanglement benchmarks | disentanglement | Helped |
| Add fade-in during progressive network growth (ablation) | pro-VLAE · Progressive training | successful training rate: 0.0 -> 0.733 | Helped |
| Add the pre-trained KL penalty for not-yet-used latent variables (ablation) | pro-VLAE · Progressive training | successful training rate: 0.0 -> 0.667 | Helped |
| Add both fade-in and pre-trained KL penalty (ablation) | pro-VLAE · Progressive training | successful training rate: 0.0 -> 0.867 | Helped |
| Use a one-dimensional latent code per layer rather than a larger code (ablation) | pro-VLAE · 3DShapes and MNIST | total I(x;z) | Hurt |
| Use four hierarchical layers with three dimensions per layer (ablation) | pro-VLAE · MNIST | clarity of information flow across layers | Hurt |
Closest papers by meaning
- Learning Disentangled Representations with Semi-Supervised Deep Generative Models (NeurIPS 2017 · similarity 0.91)
- Benchmarks, Algorithms, and Metrics for Hierarchical Disentanglement (ICML 2021 · similarity 0.91)
- Demystifying Inductive Biases for (Beta-)VAE Based Architectures (ICML 2021 · similarity 0.90)
- An Identifiable Double VAE For Disentangled Representations (ICML 2021 · similarity 0.90)
- Guided Variational Autoencoder for Disentanglement Learning (CVPR 2020 · similarity 0.90)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (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.