Research map / Optimization / Optimizers
Efficient Variance Reduction for 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 |
|---|---|---|---|
| add variance reduction (STORM-style, bilevel-aware VR-Reptile) to Reptile | Reptile · Mini-ImageNet 1-shot and 5-shot 5-way | accuracy (%): +0.4 (1-shot), +1.6 (5-shot) | Mixed |
| add variance reduction (plain STORM) to Reptile | Reptile · Mini-ImageNet 1-shot and 5-shot 5-way | accuracy (%): -0.1 (1-shot), +0.2 (5-shot) | No clear effect |
| bilevel-aware variance reduction (VR) vs STORM variant on Reptile | Reptile · Meta-Dataset 1-shot and 5-shot 5-way (Bird, Fungi) | accuracy (%): e.g. Fungi 1-shot 45.34 vs 43.56 | Helped |
| apply VR to BMG (VR-BMG) vs plain BMG | BMG · Mini-ImageNet 1-shot 5-way | accuracy (%): +0.7 | Helped |
| apply variance reduction (VFML, STORM on intermediate gradients) to Reptile | Reptile · Mini-ImageNet 1-shot 5-way | accuracy (%): -0.4 (1-shot), 0.0 vs 66.0 in 5-shot... VFML 66.2 vs Reptile 66.0 in 5-shot | Hurt |
| apply variance reduction to MAML/FOMAML (VR-MAML, VR-FOMAML) vs plain | MAML and FOMAML · Mini-ImageNet 1-shot and 5-shot 5-way | accuracy (%): MAML 48.7->49.2 (1-shot), 63.1->63.6 (5-shot); FOMAML 48.1->48.3, 63.2->63.4 | Mixed |
| generic STORM variance reduction applied to DRS (DRS+STORM) | DRS · Mini-ImageNet 1-shot 5-way | accuracy (%): +0.2 | No clear effect |
| apply bilevel variance reduction (SUSTAIN/MRBO/VRBO) to ANIL | ANIL · Mini-ImageNet 1-shot and 5-shot 5-way | accuracy (%): +0.1 to +0.3 (1-shot), +0.4 to +0.6 (5-shot) | No clear effect |
| Reptile vs K (inner steps) sweep for VR-Reptile (ablation) | VR-Reptile · Mini-ImageNet 1-shot 5-way | testing accuracy (%) | Mixed |
| use bilevel formulation (ANIL) instead of single-level meta-learning | ANIL · Mini-ImageNet 5-shot 5-way | accuracy (%): ANIL 61.4 vs Reptile 66.0 | Hurt |
| replace meta-learner with domain randomized search (DRS) | meta-learning · Mini-ImageNet 1-shot 5-way | accuracy (%): 24.5 vs 50.0 for Reptile | Hurt |
Closest papers by meaning
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.90)
- Adaptive Gradient-Based Meta-Learning Methods (NeurIPS 2019 · similarity 0.88)
- Task Agnostic Meta-Learning for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- Learning To Learn Around A Common Mean (NeurIPS 2018 · similarity 0.88)
- Submodular Meta-Learning (NeurIPS 2020 · similarity 0.87)
- Learning to Balance: Bayesian Meta-Learning for Imbalanced and Out-of-distribution Tasks (ICLR 2020 · 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.