Research map / Continual learning / Online and replay-based
Continual Learning by Asymmetric Loss Approximation With Single-Side Overestimation
Dongmin Park, Seokil Hong, Bohyung Han, Kyoung Mu Lee
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 |
|---|---|---|---|
| replace SI with ALASSO (asymmetric overestimated loss plus exact quadratic coefficient) | SI · Permuted MNIST (30 tasks) | accuracy A30: 0.752 -> 0.944 | Improved |
| replace SI with ALASSO | SI · Split CIFAR-10/CIFAR-100 | accuracy A30: 0.632 -> 0.697 | Improved |
| replace SI with ALASSO | SI · Split CIFAR-10/CIFAR-100 | forgetting F30: 0.110 -> 0.026 | Improved |
| replace SI with ALASSO | SI · Split Tiny ImageNet | average validation accuracy: approximately 10% point higher | Improved |
| replace SI with ALASSO | SI · Split CIFAR-10/CIFAR-100 | accuracy: 5.7% point better; 41% of lost accuracy recovered | Improved |
| replace SI with ALASSO | SI · Permuted MNIST (30 tasks) | intransigence I30: 0.008 -> 0.015 | Hurt |
| ALASSO versus single_task upper bound | single_task · Split Tiny ImageNet | average validation accuracy: 58.9 -> 59.4 | Improved |
| add parameter decoupling (ablation) | ALASSO · Permuted MNIST | accuracy and convergence: slight accuracy gain, faster convergence | No clear effect |
| add exact quadratic approximation of Omega (ablation) | ALASSO · Permuted MNIST | accuracy | Improved |
| add asymmetric overestimation of the unobserved side (ablation) | ALASSO · Permuted MNIST | accuracy | Improved |
| increase overestimation factor a from 1.0 (symmetric) to 2.0 (ablation) | ALASSO · Permuted MNIST (100 tasks) | accuracy: 0.62 -> 0.79 | Improved |
| increase a from 2.0 to 5.0 (ablation) | ALASSO · Permuted MNIST (100 tasks) | accuracy: 0.79 -> 0.75 | Hurt |
Closest papers by meaning
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.88)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.88)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.88)
- Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding (AAAI 2024 · similarity 0.88)
- An Effective Dynamic Gradient Calibration Method for Continual Learning (ICML 2024 · similarity 0.87)
- Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification (ICLR 2025 · similarity 0.87)
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.
