Research map / Optimization / Optimizers
Large-Scale Meta-Learning with Continual Trajectory Shifting
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 continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · Average across target datasets (1,000 training images each) | accuracy: 61.68 -> 62.12 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · CIFAR10 (1,000 training images) | accuracy: 84.58±0.12 -> 84.42±0.15 | Hurt |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · CIFAR100 (1,000 training images) | accuracy: 47.98±0.14 -> 48.34±0.21 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · SVHN (1,000 training images) | accuracy: 62.39±0.72 -> 62.82±0.56 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · Stanford Dogs (1,000 training images) | accuracy: 56.97±0.12 -> 57.53±0.48 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · VGG Pets (1,000 training images) | accuracy: 84.25±0.22 -> 84.65±0.11 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · VGG Flowers (1,000 training images) | accuracy: 87.22±0.31 -> 87.54±0.21 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · Food (1,000 training images) | accuracy: 37.35±0.22 -> 37.84±0.20 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · CUB (1,000 training images) | accuracy: 35.44±0.48 -> 36.40±0.20 | Helped |
| Use continual trajectory shifting instead of Reptile | ImageNet-pretrained ResNet18 meta-learning · DTD (1,000 training images) | accuracy: 58.98±0.59 -> 59.53±0.25 | Helped |
| Remove shifting of task-learning trajectories after meta-updates (ablation) | Continual Trajectory Shifting · VGG Pets | test accuracy | Hurt |
| Replace the meta-update shift with a shift of the same magnitude in a random direction (ablation) | Continual Trajectory Shifting · VGG Pets | test accuracy | Hurt |
Closest papers by meaning
- Continuous-Time Meta-Learning with Forward Mode Differentiation (ICLR 2022 · similarity 0.90)
- Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (ICML 2021 · similarity 0.89)
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.89)
- Meta-Learning with Adaptive Hyperparameters (NeurIPS 2020 · similarity 0.89)
- Adaptive Gradient-Based Meta-Learning Methods (NeurIPS 2019 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.88)
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.