Research map / Continual learning / Online and replay-based
Task-Free Continual Learning
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 |
|---|---|---|---|
| Add online MAS importance-weight regularization | Online learning · Breaking Bad, weak supervision | accuracy: up to a 100% improvement over the Online Baseline | Helped |
| Add online MAS importance-weight regularization | Online learning · Big Bang Theory, weak supervision | accuracy: matches the accuracy of the model trained with shuffled data; only 8% higher | Helped |
| Add online MAS importance-weight regularization | Online learning · 10-corridor simulation | accuracy: 80% vs 70% | Helped |
| Add a hard-sample buffer (ablation) | Online Baseline and Online Continual · Big Bang Theory, weak supervision | accuracy | Helped |
| Replace cumulative moving-average importance weights with a decaying average (ablation) | Online Continual · Big Bang Theory, weak supervision | accuracy stability | Hurt |
| Normalize the hard buffer to balance actions (ablation) | Online learning · 10-corridor simulation | accuracy: outperforming the baseline in all but last corridors | Helped |
| Normalize the hard buffer to balance actions (ablation) | Online learning · 10-corridor simulation | adaptation speed | Hurt |
| Add continual learning during on-policy training | Turtlebot collision avoidance · Real-world Turtlebot | collisions per gradient step: 0.6 collisions per gradient step | Helped |
| Use continual learning rather than pure online learning | Online model adapting to newly seen data · Distribution shifts with more informative new data | adaptation speed | Hurt |
| Train with a hard buffer instead of only recent samples (ablation) | Online continual learning · Synthetic binary classification | total test accuracy | Helped |
Closest papers by meaning
- Online Boundary-Free Continual Learning by Scheduled Data Prior (ICLR 2023 · similarity 0.89)
- Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime Inference (ICLR 2022 · similarity 0.89)
- Continual Learning with Adaptive Weights (CLAW) (ICLR 2020 · similarity 0.89)
- Online Bias Correction for Task-Free Continual Learning (ICLR 2023 · similarity 0.89)
- Doubly Perturbed Task Free Continual Learning (AAAI 2024 · similarity 0.88)
- Online Task-Free Continual Learning via Dynamic Expansionable Memory Distribution (CVPR 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.