Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
CMVAE: Causal Meta VAE for Unsupervised Meta-Learning
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 |
|---|---|---|---|
| Use CMVAE's causal latent-space adjustment and causal-EM instead of Meta-GMVAE | Unsupervised few-shot meta-learning · Toy dataset, 2-way 4-shot | accuracy: 78.51 ± 0.36 -> 93.08 ± 0.32 | Helped |
| Use CMVAE instead of the strongest unsupervised baseline | Unsupervised few-shot meta-learning · Omniglot | accuracy | Mixed |
| Use CMVAE instead of the strongest unsupervised baseline | Unsupervised few-shot meta-learning · miniImageNet | accuracy | Helped |
| Use CMVAE instead of IFSL | Unsupervised few-shot meta-learning · CelebA, 5-way identity recognition | accuracy | Helped |
| Use nonlinear rather than linear causal function h (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 89.26 ± 0.56 -> 95.11 ± 0.47; 42.68 ± 0.72 -> 43.91 ± 0.74; 51.28 ± 0.91 -> 59.93 ± 0.95 | Helped |
| Set λ₁ to 10⁻¹ rather than 1 (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 95.11 ± 0.47 -> 94.46 ± 0.49; 43.91 ± 0.74 -> 43.06 ± 0.75; 59.93 ± 0.95 -> 59.84 ± 0.95 | Hurt |
| Set λ₁ to 10⁵ rather than 1 (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 95.11 ± 0.47 -> 94.42 ± 0.48; 43.91 ± 0.74 -> 42.11 ± 0.74; 59.93 ± 0.95 -> 59.72 ± 0.88 | Hurt |
| Set λ₁ to 10¹⁰ rather than 1 (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 95.11 ± 0.47 -> 91.28 ± 0.61; 43.91 ± 0.74 -> 41.27 ± 0.70; 59.93 ± 0.95 -> 50.92 ± 0.92 | Hurt |
| Set λ₂ to 10⁻² rather than 10⁻⁴ (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 95.11 ± 0.47 -> 94.91 ± 0.50; 43.91 ± 0.74 -> 42.88 ± 0.75; 59.93 ± 0.95 -> 59.29 ± 0.93 | Hurt |
| Set λ₂ to 10⁻³ rather than 10⁻⁴ (ablation) | CMVAE · Omniglot, miniImageNet, and CelebA (5-way, 1-shot) | accuracy: 95.11 ± 0.47 -> 94.95 ± 0.48; 43.91 ± 0.74 -> 43.05 ± 0.75; 59.93 ± 0.95 -> 59.27 ± 0.94 | Hurt |
| Increase causal-EM over vanilla EM (ablation) | CMVAE · miniImageNet and CelebA, 5-way 1-shot | accuracy: 42.81 ± 0.70 -> 44.27 ± 0.76; 58.05 ± 0.90 -> 61.04 ± 0.94 | Helped |
Closest papers by meaning
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · similarity 0.86)
- Causal Representation Learning via Counterfactual Intervention (AAAI 2024 · similarity 0.85)
- A Meta-Learning Approach to Bayesian Causal Discovery (ICLR 2025 · similarity 0.85)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.85)
- A Sparsity Principle for Partially Observable Causal Representation Learning (ICML 2024 · similarity 0.85)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.84)
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.