Research map / Continual learning / Online and replay-based
CHEEM: Continual Learning by Reuse, New, Adapt and Skip - A Hierarchical Exploration-Exploitation Approach
Chinmay Savadikar, Michelle Dai, Tianfu Wu
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 architecture search (reuse/adapt/new/skip) over LoRA-CL (ablation) | CHEEM · MTIL (DEiT-Tiny) | average accuracy: 71.1 -> 74.5 | Improved |
| add architecture search over LoRA-CL (ablation) | CHEEM · MTIL (ViT-Base) | average accuracy: 84.7 -> 85.9 | Improved |
| use HEE sampling instead of uniform sampling in supernet training (ablation) | CHEEM · MTIL (ViT-Base) | % parameter increase: 5.89 -> 0.25 | Improved |
| use HEE sampling instead of uniform sampling in supernet training (ablation) | CHEEM · MTIL (ViT-Base) | average accuracy: 84.74 -> 85.88 | Improved |
| use HEE sampling instead of uniform sampling in supernet training (ablation) | CHEEM · MTIL (DEiT-Tiny) | average accuracy: 75.05 -> 74.51 | No clear effect |
| place CHEEM at attention projection instead of MLP Down (ablation) | CHEEM · MTIL (ViT-Base) | FLOPs: 62.31 -> 65.91 | Hurt |
| use DEiT-Tiny instead of ViT-Base for task ID inference (ablation) | CHEEM lite · MTIL (ViT-Base) | average accuracy: 85.9 -> 84.2 | No clear effect |
| use AlexNet instead of ViT-Base for task ID inference (ablation) | CHEEM lite · MTIL (ViT-Base) | average accuracy: 85.9 -> 79.9 | Hurt |
| add dynamic task-tailored backbone vs prompting baselines (S-Prompts) | CHEEM · MTIL (DEiT-Tiny) | average accuracy: 67.3 -> 74.5 | Improved |
| change task order of MTIL stream (ablation) | CHEEM · MTIL (ViT-Base) | average accuracy: 85.65, 84.74, 85.63 | No clear effect |
| use CLIP ViT-B/16 backbone | CHEEM · VDD (CLIP ViT-B/16) | average accuracy: LoRA-CL 85.8 vs CHEEM 84.5 | Mixed |
Closest papers by meaning
- MagMax: Leveraging Model Merging for Seamless Continual Learning (ECCV 2024 · similarity 0.89)
- Compacting, Picking and Growing for Unforgetting Continual Learning (NeurIPS 2019 · similarity 0.89)
- A Comprehensive Survey of Continual Learning: Theory, Method and Application (TPAMI 2024 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.87)
- Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · similarity 0.87)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 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.
