Research map / Continual learning / Online and replay-based
Addressing Catastrophic Forgetting in Few-Shot Problems
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 Bayesian online meta-learning with Laplace approximation (BomLA) instead of Sequential MAML | 5-way 1-shot triathlon sequence · Omniglot → miniQuickDraw → CIFAR-FS | meta-evaluation accuracy | Helped |
| Use Bayesian online meta-learning with variational inference (BomVI) instead of Sequential MAML | 5-way 1-shot triathlon sequence · Omniglot → miniQuickDraw → CIFAR-FS | meta-evaluation accuracy | Helped |
| Use Bayesian online meta-learning with Laplace approximation (BomLA) instead of Sequential MAML | 5-way 1-shot pentathlon sequence · Omniglot → CIFAR-FS → miniImageNet → VGG-Flowers → Aircraft | meta-evaluation accuracy | Mixed |
| Use Bayesian online meta-learning with variational inference (BomVI) instead of Sequential MAML | 5-way 1-shot pentathlon sequence · Omniglot → CIFAR-FS → miniImageNet → VGG-Flowers → Aircraft | meta-evaluation accuracy | Mixed |
| Train on each new dataset only, rather than retaining knowledge through an online posterior | Sequential MAML · 5-way 1-shot triathlon and pentathlon sequences | meta-evaluation accuracy | Mixed |
| Use Train-On-Everything (TOE), retraining from random initialization on all datasets encountered so far | TOE · 5-way 1-shot triathlon and pentathlon sequences | meta-evaluation accuracy | Helped |
| Use FTML with its modified evaluation procedure | FTML · 5-way 1-shot triathlon and pentathlon sequences | meta-evaluation accuracy | Mixed |
| Use BomLA or BomVI for sequential meta-training instead of TOE | Sequential meta-learning on Omniglot · Omniglot 5-way 5-shot novel tasks | meta-evaluation accuracy | Helped |
| Use BomLA or BomVI for sequential meta-training instead of Sequential MAML or FTML | Sequential meta-learning on Omniglot · Omniglot 5-way 5-shot novel tasks | meta-evaluation accuracy | Helped |
| Set BomLA's posterior regularizer to λ = 1000 rather than λ = 100 (ablation) | BomLA · 5-way 1-shot pentathlon sequence | meta-evaluation accuracy on previously learned datasets | Helped |
| Set BomLA's posterior regularizer to λ = 1000 rather than λ = 100 (ablation) | BomLA · 5-way 1-shot pentathlon sequence | meta-evaluation accuracy on newly encountered datasets | Hurt |
| Set BomLA's posterior regularizer to λ = 1 rather than λ = 100 (ablation) | BomLA · 5-way 1-shot triathlon and pentathlon sequences | meta-evaluation accuracy on newly encountered datasets | Helped |
Closest papers by meaning
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.90)
- Learning to Forget for Meta-Learning (CVPR 2020 · similarity 0.89)
- Look-ahead Meta Learning for Continual Learning (NeurIPS 2020 · similarity 0.89)
- Learning To Learn and Remember Super Long Multi-Domain Task Sequence (CVPR 2022 · similarity 0.89)
- Re-evaluating Continual Learning with Few-Shot Adaptation (NeurIPS 2026 · similarity 0.88)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 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.