Research map / Continual learning / Online and replay-based
SplitLoRA: Balancing Stability and Plasticity in Continual Learning Through Gradient Space Splitting
Haomiao Qiu, Miao Zhang, Ziyue Qiao, Weili Guan, Min Zhang, Liqiang Nie
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 |
|---|---|---|---|
| project A_t onto minor subspace (SplitLoRA init) instead of random init (ablation) | SplitLoRA · ImageNet-R 20-task | FAA: 72.30 -> 78.82 | Improved |
| project A_t onto minor subspace instead of InfLoRA initialization (ablation) | SplitLoRA · ImageNet-R 20-task | FAA: 73.81 -> 78.82 | Improved |
| choose per-module subspace size by stability-plasticity objective instead of fixed threshold | SplitLoRA vs InfLoRA · ImageNet-R 10-task | FAA: 78.10 -> 81.00 | Improved |
| vary alpha from 20 to 1, 5, 10 or 30 (ablation) | SplitLoRA · ImageNet-R | FAA: 80.89 - 81.38 (10-task) | No clear effect |
| increase alpha (ablation) | SplitLoRA · ImageNet-R | forgetting / plasticity | Mixed |
| swap ViT pretraining to iBOT-1K | SplitLoRA vs VPT-NSP2 · ImageNet-R 10-task | FAA: 73.85 -> 74.58 | Improved |
| swap ViT pretraining to DINO-1K | SplitLoRA vs VPT-NSP2 · ImageNet-R 10-task | FAA: 69.45 -> 70.49 | Improved |
| use SplitLoRA instead of InfLoRA on DomainNet | SplitLoRA · DomainNet 5-task | FAA: 78.26 -> 84.31 | Improved |
| use SplitLoRA instead of InfLoRA on efficiency | SplitLoRA · ImageNet-R 10-task | extra forward passes / time: 2/task -> 1/task; 1h 48m -> 1h 43m | 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.
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 7 of 7 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
- BiLoRA: Almost-Orthogonal Parameter Spaces for Continual Learning (CVPR 2025 · similarity 0.91)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.88)
- Continuous Subspace Optimization for Continual Learning (NeurIPS 2025 · similarity 0.88)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.88)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.88)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.88)
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.
