Research map / Continual learning / Online and replay-based
Reconciling meta-learning and continual learning with online mixtures of tasks
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 a mixture of meta-learners instead of a single MAML initialization | MAML · miniImageNet 5-way 1-shot | accuracy: 48.70 -> 49.60 | Helped |
| Use a mixture of meta-learners instead of a single MAML initialization | MAML · miniImageNet 5-way 5-shot | accuracy: 63.11 -> 64.60 | Helped |
| Increase the finite mixture from 2 to 3 components (ablation) | Our method · miniImageNet 5-way 1-shot | accuracy: 49.60 -> 51.20 | Helped |
| Increase the finite mixture from 2 to 3 components (ablation) | Our method · miniImageNet 5-way 5-shot | accuracy: 64.60 -> 65.00 | Helped |
| Increase the finite mixture from 3 to 4 components (ablation) | Our method · miniImageNet 5-way 1-shot | accuracy: 51.20 -> 50.49 | Hurt |
| Increase the finite mixture from 3 to 4 components (ablation) | Our method · miniImageNet 5-way 5-shot | accuracy: 65.00 -> 64.78 | Hurt |
| Increase the finite mixture from 4 to 5 components (ablation) | Our method · miniImageNet 5-way 1-shot | accuracy: 50.49 -> 51.46 | Helped |
| Use non-uniform cluster assignments rather than equal assignments (ablation) | Finite mixture meta-learner · Continual few-shot regression | mean-squared error | Helped |
| Use a Dirichlet process mixture that can add clusters online rather than a fixed three-component mixture (ablation) | Continual few-shot regression · Evolving polynomial, sinusoid, and sawtooth tasks | mean-squared error | Helped |
| Use a Dirichlet process mixture that can add clusters online rather than a fixed three-component mixture (ablation) | Continual few-shot classification · Evolving filtered miniImageNet tasks | accuracy | Helped |
| Add a cool-down period after spawning a cluster (ablation) | Task-aware nonparametric mixture · Evolving filtered miniImageNet tasks | accuracy | No clear effect |
| Train all cluster weights simultaneously with a surrogate loss rather than optimizing cluster parameters separately in alternating steps (ablation) | Finite mixture meta-learner · Reported experiments | performance | Helped |
Closest papers by meaning
- Variational Continual Bayesian Meta-Learning (NeurIPS 2021 · similarity 0.89)
- Hierarchical Gaussian Mixture based Task Generative Model for Robust Meta-Learning (NeurIPS 2023 · similarity 0.89)
- Recasting Gradient-Based Meta-Learning as Hierarchical Bayes (ICLR 2018 · similarity 0.88)
- Learning to Balance: Bayesian Meta-Learning for Imbalanced and Out-of-distribution Tasks (ICLR 2020 · similarity 0.88)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.87)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 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.