Research map / Continual learning / Online and replay-based
MaRS: Memory-Adaptive Routing for Reliable Capacity Expansion and Knowledge Retention
Gang Yan
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 SGSE statistical slot expansion (ablation) | MaRS · Tiny-ImageNet | final accuracy: ~41% | Improved |
| add anchors for knowledge retention (ablation) | MaRS · Tiny-ImageNet | final accuracy: ∼42% | Improved |
| add Stage 1 contrastive feature adaptation (ablation) | MaRS · Tiny-ImageNet | final accuracy: ∼43% | Improved |
| add Stage 2 classifier distillation (ablation) | MaRS · Tiny-ImageNet | final accuracy: ∼40% | Improved |
| initial slots S0=32 instead of 8 (ablation) | MaRS · Tiny-ImageNet | accuracy | Improved |
| enlarge initial slots S0 to 128 (ablation) | MaRS · Tiny-ImageNet | accuracy | Hurt |
| small beta=0.5 instead of 0.9 (ablation) | MaRS · Tiny-ImageNet | accuracy | Hurt |
| large beta=0.99 instead of 0.9 (ablation) | MaRS · Tiny-ImageNet | late-task accuracy | Hurt |
| unfreeze backbone (standard vs frozen-encoder) (ablation) | baselines and MaRS · CIFAR-100, Tiny-ImageNet, ASC | average accuracy: within 1–2% | No clear effect |
| replace DER++/PASS++ with MaRS | DER++ and PASS++ · CIFAR-100 and Tiny-ImageNet | average accuracy: 52–54% -> 56–58% | Improved |
| replace DER++/PASS++ with MaRS | DER++ and PASS++ · ASC | average accuracy: 74–75% -> 78–79% | Improved |
| replace CLIP-oriented baseline with MaRS | CLAP4CLIP · Tiny-ImageNet | accuracy: 46.50 -> 49.46 | Improved |
Closest papers by meaning
- Decomposing and Composing: Towards Efficient Vision-Language Continual Learning via Rank-1 Expert Pool in a Single LoRA (AAAI 2026 · similarity 0.89)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.88)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.87)
- Self-Expansion of Pre-trained Models with Mixture of Adapters for Continual Learning (CVPR 2025 · similarity 0.87)
- InduceKV: Fixed-Footprint Continual Adaptation of Multimodal LLMs via Inducing KV Memories (NeurIPS 2026 · similarity 0.86)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.86)
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.
