Research map / Continual learning / Continual fine-tuning of LLMs
DGS: Dual Gradient and Semantic-Shift Guided Low-Rank Adaptation for Class Incremental Learning
Kai Li, Jiafeng Li, Lianghua He, Ying Wen
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 Dual Gradient update (fuse raw and projected gradients) (ablation) | task-specific LoRA baseline · ImageNet-R B0Inc20 | A_last: 78.90 -> 79.68 | Improved |
| add Dual Gradient update (ablation) | task-specific LoRA baseline · ImageNet-R B0Inc20 | forgetting F_last: 9.86 -> 6.88 | Improved |
| add Dual Gradient update (ablation) | task-specific LoRA baseline · ObjectNet B0Inc10 | A_last: 61.30 -> 62.64 | Improved |
| add patch-level alignment loss on top of DG (ablation) | DGS · ImageNet-R B0Inc20 | A_last: 79.68 -> 80.15 | Improved |
| add patch-level alignment loss on top of DG (ablation) | DGS · ImageNet-R B0Inc20 | forgetting F_last: 6.88 -> 5.30 | Improved |
| add classifier alignment with semantic shift on top of DG (ablation) | DGS · ImageNet-R B0Inc20 | A_last: 79.68 -> 81.26 | Improved |
| add classifier alignment on top of DG without patch alignment (ablation) | DGS · ImageNet-R B0Inc20 | forgetting F_last: 6.88 -> 8.25 | Hurt |
| add classifier alignment on top of DG without patch alignment (ablation) | DGS · ObjectNet B0Inc10 | forgetting F_last: 10.08 -> 14.57 | Hurt |
| combine DG, patch alignment and classifier alignment (ablation) | DGS · ImageNet-R B0Inc20 | A_last: 78.90 -> 81.72 | Improved |
| combine DG, patch alignment and classifier alignment (ablation) | DGS · ObjectNet B0Inc10 | A_last: 61.30 -> 64.85 | Improved |
| increase interpolation weight alpha toward projected gradient (ablation) | Dual Gradient · ImageNet-R and ObjectNet | accuracy / forgetting | Mixed |
| vary tau scale factor from 10 to 20 to 40 (ablation) | DGS · ImageNet-R B0Inc20 | A_last: 79.15 -> 81.72 -> 79.68 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 3 of 3 shared results
- RanPAC — RanPAC: Random Projections and Pre-trained Models for Continual Learning (NeurIPS 2023) · better on 3 of 3 shared results
- SLCA — SLCA: Slow Learner with Classifier Alignment for Continual Learning on a Pre-trained Model (ICCV 2023) · better on 1 of 1 shared result
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
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · similarity 0.90)
- CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025 · similarity 0.90)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.89)
- Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning (AAAI 2025 · similarity 0.89)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.89)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · similarity 0.89)
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.
