Research map / Continual learning / Online and replay-based
SparCL: Sparse Continual Learning on the Edge
Zifeng Wang, Zheng Zhan, Yifan Gong, Geng Yuan, Wei Niu, Tong Jian, Bin Ren, Stratis Ioannidis, Yanzhi Wang, Jennifer Dy
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 |
|---|---|---|---|
| Apply SparCL with weight sparsity 0.75 to DER++ | DER++ · Split CIFAR-10 Class-IL, buffer 500 | average accuracy: 72.70 -> 74.09 | Improved |
| Apply SparCL with weight sparsity 0.75 to DER++ | DER++ · Split Tiny-ImageNet Class-IL, buffer 500 | average accuracy: 19.38 -> 20.75 | Improved |
| Use SparCL weight sparsity 0.95 with DER++ | DER++ · Split CIFAR-10 Class-IL, buffer 500 | training FLOPs (× 10^15): 13.9 -> 0.6 | Improved |
| Use SparCL instead of SNIP sparse training with DER++ | DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.90 | average accuracy: 66.07 -> 73.42 | Improved |
| Use SparCL instead of RigL sparse training with DER++ | DER++ · Split Tiny-ImageNet Class-IL, weight sparsity 0.90 | average accuracy: 15.88 -> 19.62 | Improved |
| Add task-aware dynamic masking (TDM) (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | average accuracy: 72.70 -> 73.37 | Improved |
| Add task-aware dynamic masking (TDM) (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | training FLOPs (× 10^15): 13.9 -> 3.6 | Improved |
| Add dynamic data removal (DDR) after TDM (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | average accuracy: 73.37 -> 73.80 | Improved |
| Add dynamic data removal (DDR) after TDM (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | training FLOPs (× 10^15): 3.6 -> 2.8 | Improved |
| Add dynamic gradient masking (DGM) after TDM (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | average accuracy: 73.37 -> 73.97 | Improved |
| Add dynamic gradient masking (DGM) after TDM (ablation) | SparCL-DER++ · Split CIFAR-10 Class-IL, weight sparsity 0.75 | training FLOPs (× 10^15): 3.6 -> 3.3 | Improved |
| Increase data removal proportion from 30% to at least 40% (ablation) | DDR · Split CIFAR-10 Class-IL | average accuracy | Hurt |
Closest papers by meaning
- Learning Bayesian Sparse Networks With Full Experience Replay for Continual Learning (CVPR 2022 · similarity 0.91)
- Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge (CVPR 2026 · similarity 0.89)
- Sparse Orthogonal Parameters Tuning for Continual Learning (IJCV 2026 · similarity 0.88)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.88)
- Task-Free Dynamic Sparse Vision Transformer for Continual Learning (AAAI 2024 · similarity 0.87)
- Task-aware Orthogonal Sparse Network for Exploring Shared Knowledge in Continual Learning (ICML 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.
