Research map / Continual learning / Continual fine-tuning of LLMs
Fly-CL: A Fly-Inspired Framework for Enhancing Efficient Decorrelation and Reduced Training Time in Pre-trained Model-based Continual Representation Learning
Heming Zou, Yunliang Zang, Wutong Xu, Xiangyang Ji
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 sparse random projection instead of dense (ablation) | Fly-CL (ViT-B/16) · CUB-200-2011 random projection time | seconds per task: 0.22 -> 0.08 | Improved |
| use GCV for ridge selection instead of explicit cross-validation (ablation) | Fly-CL (ViT-B/16) · CUB-200-2011 ridge selection time | seconds per task: 7.34 -> 0.14 | Improved |
| use Cholesky factorization instead of LU decomposition for prototypes (ablation) | Fly-CL (ViT-B/16) · CUB-200-2011 prototype time | seconds per task: 0.20 -> 0.10 | Improved |
| add top-k activation sparsity in similarity comparison (ablation) | Fly-CL (ViT-B/16) · CUB-200-2011 similarity time | seconds per task: 0.21 -> 0.08 | Improved |
| add the random projection layer (ablation) | Fly-CL (ViT-B/16) · CIFAR-100, CUB, VTAB accuracy curves | average accuracy | Improved |
| add streaming ridge classification (ablation) | Fly-CL (ViT-B/16) · CIFAR-100, CUB, VTAB accuracy curves | average accuracy | Improved |
| add data normalization (ablation) | Fly-CL (ViT-B/16) · CIFAR-100, CUB, VTAB accuracy curves | average accuracy | Improved |
| increase projection dimension m (ablation) | Fly-CL · CUB-200-2011 | accuracy | Improved |
| replace F-OAL with Fly-CL | F-OAL on ResNet-50 · VTAB overall accuracy | overall accuracy: 26.15 -> 94.00 | Improved |
| replace RanPAC with Fly-CL | RanPAC (ViT-B/16) · CUB-200-2011 overall accuracy | overall accuracy: 92.67 -> 93.84 | Improved |
| replace RanPAC with Fly-CL | RanPAC (ViT-B/16) · CIFAR-100 overall accuracy | overall accuracy: 94.21 -> 93.89 | Hurt |
| replace F-OAL with Fly-CL | F-OAL (ViT-B/16) · CUB-200-2011 post-extraction time | seconds per task: 2.04 -> 0.35 | Improved |
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.
- RanPAC — RanPAC: Random Projections and Pre-trained Models for Continual Learning (NeurIPS 2023) · better on 5 of 6 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
- FlyPrompt: Brain-Inspired Random-Expanded Routing with Temporal-Ensemble Experts for General Continual Learning (ICLR 2026 · similarity 0.87)
- Prototype-Sample Relation Distillation: Towards Replay-Free Continual Learning (ICML 2023 · similarity 0.87)
- Measuring Representational Shifts in Continual Learning: A Linear Transformation Perspective (ICML 2025 · similarity 0.86)
- Disentangling and mitigating the impact of task similarity for continual learning (NeurIPS 2024 · similarity 0.85)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.85)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · similarity 0.85)
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.
