Research map / Probabilistic and scientific modeling / Variational and generative sampling
Cooperation in the Latent Space: The Benefits of Adding Mixture Components in Variational 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 |
|---|---|---|---|
| Increase the mixture-component count from 1 to 4 (ablation) | Vanilla VAE · MNIST | NLL: 79.74 ± 0.03 -> 78.60 ± 0.05 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | Hierarchical VAE · MNIST | NLL: 78.92 ± 0.08 -> 77.75 ± 0.03 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | VAE w. NF · MNIST | NLL: 79.43 ± 0.03 -> 78.23 ± 0.03 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | VAE w. VampPrior · MNIST | NLL: 78.69 ± 0.02 -> 77.55 ± 0.05 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | Composite Model · MNIST | NLL: 78.45 ± 0.11 -> 77.23 ± 0.11 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | Vanilla VAE · FashionMNIST | NLL: 225.50 -> 224.22 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | Composite Model · FashionMNIST | NLL: 223.58 -> 222.38 | Helped |
| Increase the mixture-component count from 1 to 4 (ablation) | Vanilla VAE · CIFAR-10 | bits per dimension: 4.85 -> 4.83 | Helped |
| Use jointly learned mixture components instead of independently learned ensemble components | VAE latent representations · MNIST | accuracy: Ensemble VAE (S=5): 96.0±2E-3%; Mixture VAE (S=5): 96.6±2E-3% | Helped |
| Use jointly learned mixture components instead of independently learned ensemble components | VAE latent representations · FashionMNIST | accuracy: Ensemble VAE (S=5): 82.6±1E-3%; Mixture VAE (S=5): 83.5±2E-3% | Helped |
| Use more mixture components (S=1 to S=5) (ablation) | Mixture VAE using scVI · CORTEX | accuracy: 94.7±.02% -> 97.1±.01% | Helped |
| Use more mixture components (S=1 to S=5) (ablation) | Ensemble VAE using scVI · CORTEX | accuracy: 94.7±.02% -> 96.8±.01% | Helped |
Closest papers by meaning
- Hierarchical Decompositional Mixtures of Variational Autoencoders (ICML 2019 · similarity 0.90)
- Shape your Space: A Gaussian Mixture Regularization Approach to Deterministic Autoencoders (NeurIPS 2021 · similarity 0.90)
- Recursive Inference for Variational Autoencoders (NeurIPS 2020 · similarity 0.90)
- Efficient Mixture Learning in Black-Box Variational Inference (ICML 2024 · similarity 0.89)
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.88)
- A Statistical Analysis of Wasserstein Autoencoders for Intrinsically Low-dimensional Data (ICLR 2024 · similarity 0.87)
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.