Research map / Continual learning / Online and replay-based
NPCL: Neural Processes for Uncertainty-Aware Continual Learning
Saurav Jha, Dong Gong, He Zhao, Lina Yao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 hierarchical global and task-specific latent variables with distribution regularization and entropy-based head selection | NPCL · S-CIFAR-10, memory 200 | accuracy: 61.93 ± 1.79 -> 63.78 ± 1.70 | Improved |
| Use hierarchical global and task-specific latent variables with distribution regularization and entropy-based head selection | NPCL · S-CIFAR-100, memory 2000 | accuracy: 51.89 -> 46.71 | Hurt |
| Use hierarchical global and task-specific latent variables with distribution regularization and entropy-based head selection | NPCL · S-Tiny-ImageNet, memory 500 | accuracy: 17.75 ± 1.14 -> 15.29 ± 1.02 | Hurt |
| Use hierarchical global and task-specific latent variables with distribution regularization and entropy-based head selection | NPCL · P-MNIST, memory 200 | accuracy: 81.74 ± 1.07 -> 83.11 ± 0.90 | Improved |
| Use hierarchical global and task-specific latent variables with distribution regularization and entropy-based head selection | NPCL · R-MNIST, memory 500 | accuracy: 92.24 ± 1.12 -> 92.07 ± 1.39 | Hurt |
| Add global and task-specific distribution regularization (ablation) | NPCL · S-CIFAR-10 | accuracy: 32.24 -> 63.78 | Improved |
| Add task-specific distribution regularization (ablation) | NPCL · S-CIFAR-10 | accuracy: 32.24 -> 57.28 | Improved |
| Add global distribution regularization (ablation) | NPCL · S-CIFAR-10 | accuracy: 32.24 -> 50.68 | Improved |
| Remove global latent and inter-task hierarchy (ablation) | NPCL · S-CIFAR-10, memory 200 | accuracy: 63.78 ± 1.70 -> 54.6 ± 2.14 | Hurt |
| Average logits across task heads rather than selecting the lowest-entropy head (ablation) | NPCL · S-CIFAR-10, memory 200 | accuracy: 63.78 ± 1.70 -> 19.54 ± 3.44 | Hurt |
| Use NPCL rather than DER | continual learning · S-CIFAR-10, memory 200 | ECE: 0.2991 -> 0.2103 | Improved |
| Use NPCL rather than DER | few-shot replay · S-CIFAR-100, memory 5 | accuracy: 8.96 -> 10.22 | Improved |
Closest papers by meaning
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.90)
- Model Uncertainty Quantification by Conformal Prediction in Continual Learning (ICML 2025 · similarity 0.88)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.88)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.88)
- Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond (ICLR 2025 · similarity 0.88)
- A Probabilistic Framework for Modular Continual Learning (ICLR 2024 · 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.
