Research map / Continual learning / Online and replay-based
Posterior Meta-Replay for Continual Learning
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 task-specific posterior distributions instead of a single recursively updated posterior | Continual learning on SplitMNIST · SplitMNIST | TInfer final accuracy: 58.84 ± 0.64 -> 71.73 ± 0.87 | Helped |
| add parameter uncertainty to task-conditioned solutions (ablation) | PosteriorReplay on 2D mode classification · 2D mode classification | TInfer Ent final accuracy: 44.90 ± 5.74 -> 81.07 ± 6.78 | Helped |
| use an implicit rather than explicit posterior distribution (ablation) | PosteriorReplay on 2D mode classification · 2D mode classification | TInfer Ent final accuracy: 81.07 ± 6.78 -> 100.0 ± 0.00 | Helped |
| select tasks by model agreement rather than predictive entropy (ablation) | PosteriorReplay-Exp · 2D mode classification | TInfer accuracy: 81.07 ± 6.78 -> 90.02 ± 3.57 | Helped |
| infer task identity over batches instead of individual samples (ablation) | PosteriorReplay-Exp · SplitMNIST | TInfer final accuracy: 71.73 ± 0.87 -> 99.72 ± 0.02 | Helped |
| fine-tune task-specific models using stored coresets (ablation) | PosteriorReplay-Exp · SplitMNIST | TInfer final accuracy: 71.73 ± 0.87 -> 90.83 ± 0.24 | Helped |
| infer task identity over batches instead of individual samples (ablation) | PosteriorReplay-Exp · SplitCIFAR-10 | TInfer final accuracy: 61.90 ± 0.66 -> 92.94 ± 1.04 | Helped |
| fine-tune task-specific models using stored coresets (ablation) | PosteriorReplay-Exp · SplitCIFAR-10 | TInfer final accuracy: 61.90 ± 0.66 -> 64.76 ± 0.34 | Helped |
| add parameter uncertainty to task-conditioned solutions (ablation) | PosteriorReplay on SplitCIFAR-10 · SplitCIFAR-10 | TInfer final accuracy: 54.83 ± 0.79 -> 61.90 ± 0.66 | Helped |
| use an implicit rather than explicit posterior distribution (ablation) | PosteriorReplay on SplitCIFAR-10 · SplitCIFAR-10 | TInfer final accuracy: 61.90 ± 0.66 -> 51.95 ± 0.53 | Hurt |
| use a multihead output for task inference rather than a growing softmax (ablation) | EWC · SplitMNIST | TInfer final accuracy: 19.96 ± 0.07 -> 47.67 ± 1.52 | Helped |
| use a multihead output for task inference (ablation) | VCL · SplitMNIST | TInfer final accuracy | Helped |
Closest papers by meaning
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.92)
- Trust Region Continual Learning as an Implicit Meta-Learner (NeurIPS 2026 · similarity 0.90)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.89)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.89)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.88)
- BECAME: Bayesian Continual Learning with Adaptive Model Merging (ICML 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.