Research map / Continual learning / Online and replay-based
Variational Auto-Regressive Gaussian Processes 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 autoregressive conditioning between task-specific inducing variables (ablation) | VAR-GP · Split MNIST | test accuracy: 78.64 { $\pm$ 1.41} -> 90.57 { $\pm$ 1.06} | Helped |
| use autoregressive conditioning between task-specific inducing variables (ablation) | VAR-GP · Permuted MNIST | test accuracy: 96.31 { $\pm$ 0.42} -> 97.20 { $\pm$ 0.08} | Helped |
| retain variational uncertainty over kernel hyperparameters instead of using per-task MLE estimates (ablation) | VAR-GP · Split MNIST | test accuracy: 10.09 { $\pm$ 0.40} -> 90.57 { $\pm$ 1.06} | Helped |
| retain variational uncertainty over kernel hyperparameters instead of using per-task MLE estimates (ablation) | VAR-GP · Permuted MNIST | test accuracy: 10.07 -> 97.20 { $\pm$ 0.08} | Helped |
| use task-specific autoregressive inducing points instead of one global inducing-point set (ablation) | VAR-GP · Split MNIST | test accuracy: 39.31 { $\pm$ 0.28} -> 90.57 { $\pm$ 1.06} | Helped |
| use task-specific autoregressive inducing points instead of one global inducing-point set (ablation) | VAR-GP · Permuted MNIST | test accuracy: 46.02 { $\pm$ 1.09} -> 97.20 { $\pm$ 0.08} | Helped |
| use VAR-GP rather than the strongest VCL variant reported | continual learning methods · Split MNIST | test accuracy: 81.90% -> 90.57% | Helped |
| use VAR-GP rather than the strongest VCL variant reported | continual learning methods · Permuted MNIST | test accuracy: 95.06% -> 97.2% | Helped |
| retrain old inducing points during later tasks (ablation) | VAR-GP · synthetic classification dataset | retention of previous-task information | Hurt |
| add a neural feature extractor to the kernel using deep kernel learning (ablation) | VAR-GP · Split MNIST | task-wise test accuracy | Hurt |
| increase the number of inducing points (ablation) | VAR-GP · Split MNIST | mean task-wise test accuracy | Helped |
Closest papers by meaning
- Continual Learning via Sequential Function-Space Variational Inference (ICML 2022 · similarity 0.90)
- A Statistical Theory of Regularization-Based Continual Learning (ICML 2024 · similarity 0.87)
- Optimal Rates in Continual Linear Regression via Increasing Regularization (NeurIPS 2025 · similarity 0.87)
- Temporal-Difference Variational Continual Learning (NeurIPS 2025 · similarity 0.87)
- Recurrent Memory for Online Interdomain Gaussian Processes (NeurIPS 2025 · similarity 0.86)
- Bayesian Optimization via Continual Variational Last Layer Training (ICLR 2025 · similarity 0.86)
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.