Research map / Continual learning / Online and replay-based
BNS: Building Network Structures Dynamically for Continual Learning
Qi Qin, Wenpeng Hu, Han Peng, Dongyan Zhao, Bing Liu
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 |
|---|---|---|---|
| add all three task similarity measures (cos, KL, Lp) as agent input vs none (ablation) | BNS · CIFAR100 A50 | final ACC: 78.50 -> 82.39 | Improved |
| add all three task similarity measures vs none (ablation) | BNS · F-EMNIST | final ACC: 57.75 -> 60.38 | Improved |
| add all three task similarity measures vs none (ablation) | BNS · F-Celeba | final ACC: 62.28 -> 65.44 | Improved |
| add all three task similarity measures vs none (ablation) | BNS · MNIST | final ACC: 99.24 -> 99.87 | Improved |
| add all three task similarity measures vs none (ablation) | BNS · CIFAR10 | final ACC: 88.99 -> 91.40 | Improved |
| use cos+KL+Lp similarity instead of cosine only (ablation) | BNS · CIFAR100 A50 | final ACC: 80.21 -> 82.39 | Improved |
| use cos+KL+Lp similarity instead of cosine only (ablation) | BNS · F-Celeba | final ACC: 66.62 -> 65.44 | Hurt |
| use cos+KL+Lp similarity instead of cosine only (ablation) | BNS · MNIST | final ACC: 99.85 -> 99.87 | No clear effect |
| replace HAT's fixed masking with RL-built per-task network structure and initialization | HAT · F-Celeba | final ACC: 59.82 -> 65.44 | Improved |
| RL-built structure and similarity-driven initialization instead of HAT | HAT · MNIST A5 | final ACC: 99.74 -> 99.87 | No clear effect |
| BNS vs HAT on CIFAR100 A50 | HAT · CIFAR100 A50 | BWT: 0.00 -> -4.35 | Hurt |
Closest papers by meaning
- Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks (NeurIPS 2020 · similarity 0.90)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.88)
- Task-aware Orthogonal Sparse Network for Exploring Shared Knowledge in Continual Learning (ICML 2024 · similarity 0.88)
- Disentangling and mitigating the impact of task similarity for continual learning (NeurIPS 2024 · similarity 0.87)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.87)
- Do Your Best and Get Enough Rest for Continual Learning (CVPR 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.
