Research map / Continual learning / Online and replay-based
Deep Online Learning Via Meta-Learning: Continual Adaptation for Model-Based RL
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 MOLe instead of k-shot adaptation from the meta-trained prior | Model-based RL · Half-Cheetah motor malfunctions | reward | Helped |
| Use MOLe instead of continued gradient updates from the previous timestep | Model-based RL · Half-Cheetah motor malfunctions | reward | Helped |
| Use MOLe instead of standard model-based RL without online adaptation | Model-based RL · Half-Cheetah terrain slopes | reward | Helped |
| Use a meta-learned initialization for online gradient updates | Model-based RL · Half-Cheetah terrain slopes | reward | Helped |
| Use MOLe instead of continued gradient updates | Model-based RL · Six-Legged Crawler, alternating normal and crippled-leg periods | reward | Helped |
| Use MOLe instead of k-shot adaptation from the prior | Model-based RL · Six-Legged Crawler, dynamically changing leg crippling | reward | Helped |
| Use MOLe instead of model-based RL without adaptation | Model-based RL · Six-Legged Crawler, dynamically changing leg crippling | reward | Helped |
| Use MOLe instead of model-based RL with online gradient updates but no meta-training | Model-based RL · Six-Legged Crawler, dynamically changing leg crippling | reward | Helped |
| Add task-specific mixture assignment and recall to continuous gradient adaptation | Continued adaptation with meta-learning · Six-Legged Crawler, walking forward in a normal configuration | performance: 200% lower | Helped |
| Add EM-weighted task updates during MAML meta-training (ablation) | MOLe · Model-based RL test tasks | performance | No clear effect |
| Train the MAML prior with more data (ablation) | MOLe · Unseen test tasks | performance | Helped |
Closest papers by meaning
- Learning to Adapt in Dynamic, Real-World Environments through Meta-Reinforcement Learning (ICLR 2019 · similarity 0.90)
- Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments (ICLR 2018 · similarity 0.90)
- Online Meta-Learning (ICML 2019 · similarity 0.89)
- Continual Slow-and-Fast Adaptation of Latent Neural Dynamics (CoSFan): Meta-Learning What-How & When to Adapt (ICLR 2025 · similarity 0.89)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.89)
- Online Learning via Learned Latent Bayesian Tracking (NeurIPS 2026 · 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.