Research map / Continual learning / Online and replay-based
Elastic Multi-Gradient Descent for Parallel Continual Learning
Fan Lyu, Wei Feng, Yuepan Li, Qing Sun, Fanhua Shang, Liang Wan, Liang Wang
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 gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-CIFAR-100 (task-incremental) | final average accuracy: 52.798 -> 56.611 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-ImageNet-TINY (task-incremental) | final average accuracy: 46.666 -> 48.746 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-EMNIST (task-incremental) | final average accuracy: 90.442 -> 90.402 | No clear effect |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-EMNIST (class-incremental) | final average accuracy: 53.441 -> 56.891 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-CIFAR-100 (class-incremental) | final average accuracy: 13.705 -> 16.362 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-CIFAR-100 task-incremental | forgetting F_e (%): 26.858 -> 23.970 | Mixed |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · PS-CIFAR-100 20-epoch (class-incremental) | A_first (%): 26.984 -> 30.657 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · Blurring PCL CIFAR-100 (task-incremental) | A_first (%): 34.860 -> 45.139 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GS) · Blurring PCL CIFAR-100 (class-incremental) | A_first (%): 12.861 -> 18.912 | Improved |
| add gradient-guided memory editing (ME) (ablation) | EMGD(GMC) · Split-CIFAR-100 (SCL, online) | final average accuracy: 47.037 -> 49.500 | Improved |
| use elastic factors (EMGD GMC) instead of equal weights (MGDA) | MaxDO/MGDA baseline · PS-ImageNet-TINY (task-incremental) | final average accuracy: 40.770 -> 46.666 | Improved |
| use EMGD(GS) instead of AvgGrad in MTL | MTL multi-task learning · Split-CIFAR-100 (MTL) | accuracy: 23.303 -> 24.780 | Improved |
Closest papers by meaning
- Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification (ICLR 2025 · similarity 0.88)
- An Effective Dynamic Gradient Calibration Method for Continual Learning (ICML 2024 · similarity 0.88)
- Pareto Continual Learning: Preference-Conditioned Learning and Adaption for Dynamic Stability-Plasticity Trade-off (AAAI 2025 · similarity 0.87)
- Continuous Subspace Optimization for Continual Learning (NeurIPS 2025 · similarity 0.87)
- Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning (ICLR 2024 · similarity 0.87)
- A Statistical Theory of Regularization-Based 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.
