Research map / Continual learning / Online and replay-based
Variational 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 online variational inference instead of tuned continual-learning regularization baselines | continual learning on Permuted MNIST · Permuted MNIST, after 10 tasks | average test accuracy: 90% vs 84%, 86%, and 82% | Helped |
| add a coreset to VCL (ablation) | VCL · Permuted MNIST, after 10 tasks | average test accuracy: 90% -> 93% | Helped |
| increase random coreset size to 5,000 examples per task (ablation) | VCL · Permuted MNIST, after 10 tasks | average test accuracy: 90% -> 95.5% | Helped |
| use online variational inference instead of EWC or diagonal LP | continual learning on Split MNIST · Split MNIST, after 5 tasks | average test accuracy: 97.0% vs 63.1% and 61.2% | Helped |
| use online variational inference instead of SI | continual learning on Split MNIST · Split MNIST, after 5 tasks | average test accuracy: 97.0% vs 98.9% | Hurt |
| add a coreset (ablation) | VCL · Split MNIST, after 5 tasks | average test accuracy: 97.0% -> around 98.4% | Helped |
| use online variational inference instead of EWC or diagonal LP | continual learning on Split notMNIST · Split notMNIST, after 5 tasks | average test accuracy: 92.0% vs 71% and 63% | Helped |
| use online variational inference instead of SI | continual learning on Split notMNIST · Split notMNIST, after 5 tasks | average test accuracy: 92.0% vs 94% | Hurt |
| add a random coreset (ablation) | VCL · Split notMNIST, after 5 tasks | average test accuracy: 92.0% -> 96% | Helped |
| use VCL rather than EWC | continual learning for deep generative models · MNIST and notMNIST generation | test log-likelihood | Mixed |
| replace continual-learning updates with naïve online VAE training (ablation) | deep generative model · MNIST and notMNIST generation | generated sample quality | Hurt |
Closest papers by meaning
- Temporal-Difference Variational Continual Learning (NeurIPS 2025 · similarity 0.89)
- Provable Contrastive Continual Learning (ICML 2024 · similarity 0.86)
- A Statistical Theory of Regularization-Based Continual Learning (ICML 2024 · similarity 0.86)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.86)
- GACL: Exemplar-Free Generalized Analytic Continual Learning (NeurIPS 2024 · similarity 0.85)
- KAC: Kolmogorov-Arnold Classifier for Continual Learning (CVPR 2025 · similarity 0.85)
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.