Research map / Continual learning / Class-incremental
Random Anchors with Low-rank Decorrelated Learning: A Minimalist Pipeline for Class-Incremental Medical Image Classification
Xinyao Wu, Zhe Xu, Raymond Kai-yu Tong
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 first-stage adaptation (ablation) | RA-LDL (with LRP) · COVID (CT&X-rays) | AccLast (%): 90.30 -> 88.04 | Hurt |
| add first-stage adaptation (ablation) | RA-LDL (with LRP) · Skin8 | AccLast (%): 60.85 -> 62.49 | Improved |
| add first-stage adaptation (ablation) | RA-LDL (with LRP) · MedMNIST-Sub | AccLast (%): 70.29 -> 70.60 | No clear effect |
| add LRP residual to random anchor (without first-stage adaptation) (ablation) | RA-DL · COVID (CT&X-rays) | AccLast (%): 86.94 -> 90.30 | Improved |
| add LRP residual to random anchor (without first-stage adaptation) (ablation) | RA-DL · Skin8 | AccLast (%): 61.26 -> 60.85 | Hurt |
| add LRP residual to random anchor (without first-stage adaptation) (ablation) | RA-DL · MedMNIST-Sub | AccLast (%): 69.32 -> 70.29 | Improved |
| add analytical decorrelated classifier instead of prototype classifier (ablation) | ViT-B/16-IN21K classifier · MedMNIST-Sub | accuracy: 50.45 -> 66.44 | Improved |
| add analytical decorrelated classifier instead of prototype classifier (ablation) | ViT-B/16-IN21K classifier · IN-A | accuracy: 48.39 -> 54.18 | Improved |
| use standard Gaussian instead of Kaiming uniform for random matrix (ablation) | RA-LDL w/o first-stage adaptation · Blood | AccLast (%): 83.71 -> 97.74 | Improved |
| use standard Gaussian instead of Kaiming uniform with parameter 5 (ablation) | RA-LDL w/o first-stage adaptation · Blood | AccLast (%): 19.31 -> 97.74 | Improved |
| increase random projection dimension from d0 to 5d0 (ablation) | RA-LDL w/o first-stage adaptation · CIFAR | AccLast (%): 85.61 -> 87.63 | Improved |
| increase random projection dimension from d0 to 5d0 (ablation) | RA-LDL w/o first-stage adaptation · Skin8 | AccLast (%): 56.74 -> 60.85 | Improved |
Closest papers by meaning
- DK-DDIL: Adaptive Knowledge Retention for Dynamic Domain-Incremental Learning in Medical Imaging (CVPR 2026 · similarity 0.87)
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · similarity 0.85)
- Samba: Severity-aware Recurrent Modeling for Cross-domain Medical Image Grading (NeurIPS 2024 · similarity 0.85)
- TinyMIG: Transferring Generalization from Vision Foundation Models to Single-Domain Medical Imaging (ICML 2025 · similarity 0.85)
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.85)
- Memory-Efficient Prompt Tuning for Incremental Histopathology Classification (AAAI 2024 · similarity 0.84)
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.
