Research map / Continual learning / Continual fine-tuning of LLMs
RanPAC: Random Projections and Pre-trained Models for Continual Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the frozen nonlinear random projection layer (ablation) | RanPAC · CIFAR100 | Average Accuracy: 90.6% -> 92.2% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · ImageNet-R | Average Accuracy: 73.9% -> 77.9% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · ImageNet-A | Average Accuracy: 57.7% -> 62.4% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · CUB | Average Accuracy: 86.6% -> 90.3% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · OmniBenchmark | Average Accuracy: 74.3% -> 79.9% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · VTAB | Average Accuracy: 90.0% -> 92.2% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · Stanford Cars | Average Accuracy: 73.0% -> 77.5% | Helped |
| add first-session PETL training (ablation) | RanPAC · Stanford Cars | Average Accuracy: 66.0% -> 77.5% | Helped |
| add first-session PETL training (ablation) | RanPAC · ImageNet-A | Average Accuracy: 58.2% -> 62.4% | Helped |
| add first-session PETL training (ablation) | RanPAC · VTAB | Average Accuracy: 92.2% -> 92.2% | No clear effect |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · DomainNet | Average Accuracy: 55.2% -> 66.6% | Helped |
| add the frozen nonlinear random projection layer (ablation) | RanPAC · CORe50 | Average Accuracy: 94.8% -> 96.7% | Helped |
Closest papers by meaning
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.92)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.90)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.90)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.90)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.90)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.90)
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.