Research map / Continual learning / Online and replay-based
Continual Learning through Control Minimization
Sander de Haan, Yassine Taoudi-Benchekroun, Pau Vilimelis Aceituno, Benjamin F Grewe
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 EFC control-minimization instead of oEWC | regularization-based continual learning · Split-CIFAR10 (Class-IL) | accuracy (%): 21.2 -> 50.2 | Improved |
| use EFC instead of EWC | Split-MNIST MLP · Split-MNIST (Class-IL) | accuracy (%): 19.8 -> 51.4 | Improved |
| use EFC instead of oEWC | Split-Tiny-ImageNet · Split-Tiny-ImageNet (Class-IL) | accuracy (%): 4.8 -> 8.8 | Improved |
| use EFC instead of oEWC in task-incremental learning | Split-MNIST · Split-MNIST (Task-IL) | accuracy (%): 99.2 -> 98.3 | No clear effect |
| use EFC instead of EWC in task-incremental learning | Split-CIFAR10 · Split-CIFAR10 (Task-IL) | accuracy (%): 95.5 -> 96.2 | No clear effect |
| add replay (DER++, buffer 200) rather than EFC | Split-MNIST class-IL · Split-MNIST (Class-IL) | accuracy (%): 51.4 -> 71.1 | Improved |
| replace diagonal Fisher (EWC) with full Fisher (FISH) in regularization | EWC on two-task Split-MNIST · Split-MNIST two-task class-IL (MLP 64x2) | Task-A forgetting | Improved |
| use EFC instead of full-Fisher FISH | two-task Split-MNIST · Split-MNIST two-task class-IL (MLP 64x2) | Task-A accuracy decrease | Improved |
| improve curvature approximation (diagonal to full Fisher) in backprop regularization | class-incremental learning · Split-MNIST two-task class-IL (MLP 64x2) | class-IL loss convergence | No clear effect |
Closest papers by meaning
- Understanding the Dynamics of Forgetting and Generalization in Continual Learning via the Neural Tangent Kernel (ICLR 2026 · similarity 0.91)
- Understanding Forgetting in Continual Learning with Linear Regression (ICML 2024 · similarity 0.91)
- On the Theory of Continual Learning with Gradient Descent for Neural Networks (ICML 2026 · similarity 0.90)
- Optimal Continual Learning has Perfect Memory and is NP-hard (ICML 2020 · similarity 0.90)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.89)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · similarity 0.89)
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.
