Research map / Continual learning / Online and replay-based
Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks
Zixuan Ke, Bing Liu, Xingchang Huang
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 CAT with task masks, similarity detection, and KTA | continual learning of mixed task sequences · Four mixed-sequence datasets, 2-layer fully connected network | overall accuracy: 0.7293 vs 0.7337; 0.6194 vs 0.5967; 0.8439 vs 0.8245; 0.6843 vs 0.6796 | Improved |
| Use CAT with task masks, similarity detection, and KTA | continual learning of mixed task sequences · M(CIFAR100-10, F-CelebA) and M(CIFAR100-20, F-CelebA), AlexNet-like architecture | overall accuracy: 0.6764 -> 0.6831; 0.7428 -> 0.7468 | Improved |
| Add backward knowledge transfer to forward transfer (ablation) | CAT · F-EMNIST and F-CelebA tasks in four mixed sequences | accuracy: 0.6104 -> 0.6134; 0.6081 -> 0.6187; 0.6873 -> 0.6909; 0.6782 -> 0.7164 | Improved |
| Add KTA (ablation) | CAT · Overall accuracy across four mixed-sequence datasets | accuracy: 0.7585 -> 0.7710; 0.5915 -> 0.6194; 0.8423 -> 0.8439; 0.6572 -> 0.6843 | Improved |
| Detect task similarity instead of treating all previous tasks as dissimilar (ablation) | CAT · Overall accuracy across four mixed-sequence datasets | accuracy: 0.7337 -> 0.7710; 0.5967 -> 0.6194; 0.8245 -> 0.8439; 0.6796 -> 0.6843 | Improved |
| Detect task similarity instead of treating all previous tasks as similar (ablation) | CAT · Overall accuracy across four mixed-sequence datasets | accuracy: 0.1887 -> 0.7710; 0.3090 -> 0.6194; 0.3631 -> 0.8439; 0.3207 -> 0.6843 | Improved |
| Add task masks and similarity detection to the all-sim, no-KTA variant (ablation) | CAT · Overall accuracy across four mixed-sequence datasets | accuracy: 0.7293 -> 0.7710; 0.5535 -> 0.6194; 0.8024 -> 0.8439; 0.6018 -> 0.6843 | Improved |
| Add KTA (ablation) | CAT · Dissimilar-task subsets in mixed sequences | accuracy: 0.9301 -> 0.9287; 0.5594 -> 0.5479; 0.9637 -> 0.9566; 0.6740 -> 0.6683 | Mixed |
| Transfer selectively from similar prior tasks using KTA (ablation) | CAT · Similar-task subsets in four mixed sequences | accuracy: 0.5870 -> 0.6134; 0.6236 -> 0.6909; 0.5995 -> 0.6187; 0.6236 -> 0.7164 | Improved |
Closest papers by meaning
- Disentangling and mitigating the impact of task similarity for continual learning (NeurIPS 2024 · similarity 0.89)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.88)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.88)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.87)
- The Joint Effect of Task Similarity and Overparameterization on Catastrophic Forgetting — An Analytical Model (ICLR 2024 · similarity 0.87)
- Continual Learning with Global Alignment (NeurIPS 2024 · 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.
