Research map / Probabilistic and scientific modeling / Variational and generative sampling
Meta-Learning with Latent Embedding Optimization
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 |
|---|---|---|---|
| Optimize generated parameters in a data-conditioned latent space | LEO · miniImageNet 1-shot and 5-shot; tieredImageNet 1-shot and 5-shot | accuracy: 61.76 \pm 0.08\% vs 58.50 \pm 0.30\%; 77.59 \pm 0.12\% vs 76.70 \pm 0.30 \%; 66.33 \pm 0.05\% vs 62.95 \pm 0.03\%; 81.44 | Helped |
| Use multi-view feature representations | LEO · miniImageNet 1-shot and 5-shot | accuracy: 63.97 \pm 0.20\%; 79.49\pm 0.70 \% | Helped |
| Add latent-space adaptation to the conditional generator (ablation) | Conditional parameter generator · miniImageNet and tieredImageNet, 1-shot and 5-shot | accuracy: 60.33 \pm 0.11\% -> 61.76 \pm 0.08\%; 74.53 \pm 0.11\% -> 77.59 \pm 0.12\%; 65.17\pm 0.15\% -> 66.33 \pm 0.05\%; 78.77\p | Helped |
| Add latent-space adaptation rather than relying on parameter-space fine-tuning (ablation) | Conditional generator + fine-tuning · miniImageNet and tieredImageNet, 1-shot and 5-shot | accuracy: 60.62 \pm 0.31\% -> 61.76 \pm 0.08\%; 76.42 \pm 0.09\% -> 77.59 \pm 0.12\%; 65.74 \pm 0.28\% -> 66.33 \pm 0.05\%; 80.65 | Helped |
| Add data-conditioned encoding instead of using the random prior (ablation) | LEO · miniImageNet and tieredImageNet, 1-shot and 5-shot | accuracy: 61.01 \pm 0.12\% -> 61.76 \pm 0.08\%; 77.27 \pm 0.05\% -> 77.59 \pm 0.12\%; 65.39\pm 0.10\% -> 66.33 \pm 0.05\%; 80.83\p | Helped |
| Add stochastic sampling rather than use deterministic codes (ablation) | LEO · miniImageNet 1-shot and 5-shot | accuracy: 61.48 \pm 0.05\% -> 61.76 \pm 0.08\%; 76.53 \pm 0.24\% -> 77.59 \pm 0.12\% | Mixed |
| Add stochastic sampling rather than use deterministic codes (ablation) | LEO · tieredImageNet 1-shot and 5-shot | accuracy: 66.18 \pm 0.17\% -> 66.33 \pm 0.05\%; 82.06\pm 0.08\% -> 81.44\pm 0.09\% | Hurt |
| Add parameter-space fine-tuning after latent adaptation (ablation) | LEO · miniImageNet 1-shot and 5-shot; tieredImageNet 1-shot | accuracy: 61.62 \pm 0.15\% -> 61.76 \pm 0.08\%; 77.46 \pm 0.12\% -> 77.59 \pm 0.12\%; 66.14 \pm 0.17\% -> 66.33 \pm 0.05\% | No clear effect |
| Add parameter-space fine-tuning after latent adaptation (ablation) | LEO · tieredImageNet 5-shot | accuracy: 80.89\pm 0.11\% -> 81.44\pm 0.09\% | Helped |
| Use a low-dimensional latent bottleneck instead of direct parameter-space adaptation | Meta-SGD versus LEO · miniImageNet and tieredImageNet, 1-shot and 5-shot | accuracy: 54.24 \pm 0.03\% -> 61.76 \pm 0.08\%; 70.86 \pm 0.04\% -> 77.59 \pm 0.12\%; 62.95 \pm 0.03\% -> 66.33 \pm 0.05\%; 79.34 | Helped |
Closest papers by meaning
- On Enforcing Better Conditioned Meta-Learning for Rapid Few-Shot Adaptation (NeurIPS 2022 · similarity 0.89)
- Meta-Learning With Differentiable Convex Optimization (CVPR 2019 · similarity 0.88)
- Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models (ICLR 2021 · similarity 0.88)
- MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning (AAAI 2024 · similarity 0.87)
- Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 2025 · similarity 0.86)
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.