Research map / Continual learning / Online and replay-based
iTAML: An Incremental Task-Agnostic Meta-learning Approach
Check 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 task-specific classifier weights and the task-balanced meta-update instead of joint training or a generic shared classifier | iTAML · CIFAR100 (10 tasks, task-agnostic) | classification accuracy | Helped |
| Use iTAML instead of RPS-net | class-incremental learning · MNIST (A5) | accuracy: 96.16% -> 97.95% | Helped |
| Use iTAML instead of RPS-net | class-incremental learning · SVHN (A5) | accuracy: 88.91% -> 93.97% | Helped |
| Use iTAML instead of RPS-net | class-incremental learning · ImageNet-100/10 | accuracy: +15.7% | Helped |
| Use iTAML instead of iCaRL | class-incremental learning · ImageNet-1K/10 | accuracy: +19.1% | Helped |
| Use iTAML instead of BiC | class-incremental learning · MS-Celeb-10K/10 | accuracy: +7.4% | Helped |
| Increase the number of inner-loop updates from r=1 to r=5 (ablation) | iTAML · CIFAR100 (10 tasks) | accuracy: 77.79% -> 81.57% | Helped |
| Increase beta from 0 toward 2 (ablation) | iTAML · CIFAR100 (10 tasks) | accuracy | Helped |
| Increase beta beyond 2 (ablation) | iTAML · CIFAR100 (10 tasks) | accuracy | Hurt |
| Increase inference data-continuum size p (ablation) | iTAML · CIFAR100 (10 tasks) | task prediction accuracy | Helped |
| Increase adaptation epochs from b=1 to b=5 (ablation) | iTAML · CIFAR100 (2 tasks) | accuracy: 78.24%, 78.48%, 78.48%, 78.53%, 78.50% | No clear effect |
| Use a smaller 0.49M-parameter backend rather than full ResNet-18 (ablation) | iTAML · CIFAR100 | accuracy: 80.27% -> 69.94% | Hurt |
Closest papers by meaning
- MetaFSCIL: A Meta-Learning Approach for Few-Shot Class Incremental Learning (CVPR 2022 · similarity 0.89)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.89)
- Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (ICML 2021 · similarity 0.88)
- MIND: Multi-Task Incremental Network Distillation (AAAI 2024 · similarity 0.87)
- Task-Agnostic Guided Feature Expansion for Class-Incremental Learning (CVPR 2025 · similarity 0.87)
- Class Incremental Learning via Likelihood Ratio Based Task Prediction (ICLR 2024 · 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.