Research map / Continual learning / Continual fine-tuning of LLMs
Forging a Dynamic Memory: Retrieval-Guided Continual Learning for Generalist Medical Foundation Models
Zizhi Chen, Yizhen Gao, Minghao Han, Yizhou Liu, Zhaoyu Chen, Dingkang Yang, Lihua Zhang
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 |
|---|---|---|---|
| Use last-task model as distillation teacher instead of initial CLIP (ablation) | PRIMED · HieraMedTransfer Order I | Last: 78.2 -> 82.1 | Improved |
| Use contrastive (image+text) distillation instead of image-only (ablation) | PRIMED · HieraMedTransfer Order I | Last: 81.3 -> 82.1 | Improved |
| Use contrastive distillation instead of text-only (ablation) | PRIMED · HieraMedTransfer Order I | Last: 80.5 -> 82.1 | Improved |
| Use Dynamic Fisher Weight Guard instead of l2 (ablation) | PRIMED · HieraMedTransfer Order I | Transfer: 57.2 -> 58.3 | Improved |
| Use Dynamic Fisher Weight Guard instead of EWC (ablation) | PRIMED · HieraMedTransfer Order I | Avg: 71.3 -> 73.1 | Improved |
| Use hierarchical retrieval instead of BM25 (ablation) | PRIMED · HieraMedTransfer Order I | Transfer: 55.1 -> 58.3 | Improved |
| Use hierarchical retrieval instead of embedding-only (ablation) | PRIMED · HieraMedTransfer Order I | Transfer: 56.1 -> 58.3 | Improved |
| Add CMC loss (beta 0 -> 0.25) (ablation) | PRIMED · HieraMedTransfer Order I | Transfer / Last: Transfer 57.2 -> 58.3; Last 82.5 -> 82.1 | Mixed |
| Use WISE-weighted teacher instead of last-task teacher (ablation) | PRIMED · HieraMedTransfer Order I | Last: 82.1 -> 79.1 | Hurt |
| Dynamic multi-stage retrieval reference data instead of uniform ZSCL reference data | PRIMED · HieraMedTransfer Order II | Last: 77.0 -> 81.2 | Improved |
| Dynamic retrieval vs uniform reference data | PRIMED · MedXtreme Order II | ACC: 64.5 -> 68.1 | Improved |
| Apply LwF baseline | LwF · HieraMedTransfer Order I | Avg: 68.5 -> 54.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Continual Self-supervised Learning: Towards Universal Multi-modal Medical Data Representation Learning (CVPR 2024 · similarity 0.88)
- HealthGPT: A Medical Large Vision-Language Model for Unifying Comprehension and Generation via Heterogeneous Knowledge Adaptation (ICML 2025 · similarity 0.88)
- SMMILE: An expert-driven benchmark for multimodal medical in-context learning (NeurIPS 2025 · similarity 0.88)
- MMed-RAG: Versatile Multimodal RAG System for Medical Vision Language Models (ICLR 2025 · similarity 0.88)
- Uni-Med: A Unified Medical Generalist Foundation Model For Multi-Task Learning Via Connector-MoE (NeurIPS 2024 · similarity 0.88)
- Advancing Multiple Instance Learning with Continual Learning for Whole Slide Imaging (CVPR 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.
