Research map / Continual learning / Online and replay-based
Continual Learning From a Stream of APIs
Enneng Yang, Zhenyi Wang, Li Shen, Nan Yin, Tongliang Liu, Guibing Guo, X. Wang, Dacheng Tao
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 memory replay loss (ablation) | DFCL-APIs · NLP 5-dataset sequence | accuracy (%): 58.49 -> 66.20 | Improved |
| add network similarity loss (ablation) | DFCL-APIs · NLP 5-dataset sequence | accuracy (%): 64.45 -> 66.20 | Improved |
| add both memory replay and network similarity losses (ablation) | DFCL-APIs · NLP 5-dataset sequence | accuracy (%): 54.68 -> 66.20 | Improved |
| provide 2% raw data (DECL-APIs instead of DFCL-APIs) | DFCL-APIs · NLP 5-dataset sequence | accuracy (%): 66.20 -> 72.62 | Improved |
| provide 2% raw data (DECL-APIs instead of DFCL-APIs) | DFCL-APIs · Split-SVHN | accuracy (%): 94.29 -> 96.01 | Improved |
| provide 10% raw data (DECL-APIs instead of DFCL-APIs) | DFCL-APIs · Split-CIFAR100 | accuracy (%): 30.52 -> 51.90 | Improved |
| provide 10% raw data (DECL-APIs instead of DFCL-APIs) | DFCL-APIs · Split-MiniImageNet | accuracy (%): 22.95 -> 51.49 | Improved |
| use DECL-APIs with 2% data instead of Classic CL with 2% data | Classic CL · Split-SVHN | accuracy (%): 55.14 -> 96.01 | Improved |
| use DECL-APIs with 2% data instead of Classic CL with 2% data | Classic CL · NLP 5-dataset sequence | accuracy (%): 27.96 -> 72.62 | Improved |
| return only hard labels from the API instead of soft probabilities | DFCL-APIs · Split-MNIST | accuracy (%): 98.58 -> 95.29 | Hurt |
| return only hard labels from the API instead of soft probabilities | DECL-APIs (2%) · Split-MNIST | accuracy (%): 98.60 -> 97.65 | Hurt |
| black-box API access instead of white-box model access | Ex-Model · Split-CIFAR100 | accuracy (%): 36.98 -> 30.52 | Hurt |
Closest papers by meaning
- CLDyB: Towards Dynamic Benchmarking for Continual Learning with Pre-trained Models (ICLR 2025 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.86)
- Doubly Perturbed Task Free Continual Learning (AAAI 2024 · similarity 0.86)
- TinySubNets: An Efficient and Low Capacity Continual Learning Strategy (AAAI 2025 · similarity 0.86)
- An Effective Dynamic Gradient Calibration Method for Continual Learning (ICML 2024 · similarity 0.85)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.85)
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.
