Research map / Image generation / Autoregressive and latent generation
Hierarchical Quantized Autoencoders
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 hierarchy (HAMs) over single-layer VQ-VAE (ablation) | VQ-VAE · MNIST 8-bit 1x1 | rFID: 85.9 -> 45.6 | Helped |
| add hierarchy (HAMs) over single-layer VQ-VAE (ablation) | VQ-VAE · MNIST 8-bit 1x1 | Distortion (MSE): 0.040 -> 0.090 | Hurt |
| add Gumbel Softmax and code resetting to HAMs (ablation) | HAMs · MNIST 8-bit 1x1 | rFID: 45.6 -> 38.6 | Helped |
| add Gumbel Softmax and code resetting to HAMs (ablation) | HAMs · MNIST 8-bit 1x1 | Classification error (%): 29.9 -> 51.1 | Hurt |
| use MSE loss on all layers of HAMs (ablation) | HAMs · MNIST 8-bit 1x1 | rFID: 45.6 -> 36.0 | Helped |
| use MSE loss on all layers of HAMs (ablation) | HAMs · MNIST 8-bit 1x1 | Classification error (%): 29.9 -> 11.1 | Helped |
| combine Gumbel Softmax with MSE loss (ablation) | HAMs · MNIST 8-bit 1x1 | rFID: 36.0 -> 21.0 | Helped |
| add probabilistic loss to HAMs+GS+MSE (ablation) | HAMs + GS + MSE · MNIST 8-bit 1x1 | rFID: 21.0 -> 22.8 | Hurt |
| add probabilistic loss to HAMs+GS+MSE (ablation) | HAMs + GS + MSE · MNIST 8-bit 1x1 | Classification error (%): 10.6 -> 12.4 | Hurt |
Closest papers by meaning
- Variational Bayesian Quantization (ICML 2020 · similarity 0.91)
- Improving Inference for Neural Image Compression (NeurIPS 2020 · similarity 0.91)
- Restructuring Vector Quantization with the Rotation Trick (ICLR 2025 · similarity 0.89)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.89)
- Finite Scalar Quantization: VQ-VAE Made Simple (ICLR 2024 · similarity 0.88)
- Robustly overfitting latents for flexible neural image compression (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.