Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models
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 |
|---|---|---|---|
| Generate synthetic tasks by latent-space interpolation with random out-of-class samples | LASIUM-MAML · Omniglot 5-way 1-shot | accuracy: 81.91 -> 83.26 | Helped |
| Generate synthetic tasks by latent-space interpolation with random out-of-class samples | LASIUM-MAML · CelebA identity 5-way 5-shot | accuracy: 62.71 -> 66.98 | Helped |
| Use latent-space noise sampling with a pretrained generator | LASIUM-N-GAN-MAML · CelebA attributes 2-way 5-shot | accuracy: 74.98 ± 1.02 vs 74.79 ± 1.01 | No clear effect |
| Use latent-space noise sampling to generate tasks | LASIUM-N-GAN-MAML · mini-ImageNet 5-way 50-shot | accuracy: 69.64 -> 69.13 | Hurt |
| Increase LASIUM-N noise variance from σ²=0.5 to σ²=1.0 (ablation) | LASIUM-N-GAN-MAML · Omniglot, 1000 iterations | accuracy: 77.16 ± 0.65 -> 71.10 ± 0.70 | Hurt |
| Increase LASIUM-N noise variance from σ²=1.0 to σ²=2.0 (ablation) | LASIUM-N-GAN-MAML · Omniglot, 1000 iterations | accuracy: 71.10 ± 0.70 -> 63.18 ± 0.71 | Hurt |
| Increase LASIUM-RO interpolation ratio from α=0.2 to α=0.4 (ablation) | LASIUM-RO-GAN-MAML · Omniglot, 1000 iterations | accuracy: 77.62 ± 0.64 -> 75.79 ± 0.65 | Hurt |
| Replace random out-of-class interpolation with interpolation toward other task classes at α=0.2 (ablation) | LASIUM-GAN-MAML · Omniglot, 1000 iterations | accuracy: 77.62 ± 0.64 -> 74.70 ± 0.68 | Hurt |
| Replace random out-of-class interpolation with interpolation toward other task classes at α=0.4 (ablation) | LASIUM-VAE-MAML · Omniglot, 1000 iterations | accuracy: 71.31 ± 0.70 -> 68.79 ± 0.73 | Hurt |
| Increase LASIUM-RO interpolation ratio from α=0.2 to α=0.4 (ablation) | LASIUM-RO-GAN-ProtoNets · Omniglot, 1000 iterations | accuracy: 62.24 ± 0.79 -> 64.19 ± 0.76 | Helped |
| Increase LASIUM-RO interpolation ratio from α=0.2 to α=0.4 (ablation) | LASIUM-RO-GAN-ProtoNets · Omniglot, 1000 iterations | accuracy: 61.79 ± 0.79 -> 64.59 ± 0.76 | Helped |
| Increase LASIUM-RO interpolation ratio from α=0.2 to α=0.4 (ablation) | LASIUM-RO-VAE-MAML · Omniglot, 1000 iterations | accuracy: 75.02 ± 0.66 -> 71.31 ± 0.70 | Hurt |
Closest papers by meaning
- Unsupervised Learning via Meta-Learning (ICLR 2019 · similarity 0.89)
- Unsupervised Meta-learning via Few-shot Pseudo-supervised Contrastive Learning (ICLR 2023 · similarity 0.89)
- Meta-GMVAE: Mixture of Gaussian VAE for Unsupervised Meta-Learning (ICLR 2021 · similarity 0.89)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · similarity 0.88)
- Meta-Learning with Latent Embedding Optimization (ICLR 2019 · similarity 0.88)
- Set-based Meta-Interpolation for Few-Task Meta-Learning (NeurIPS 2022 · 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.