Research map / Continual learning / Continual fine-tuning of LLMs
COLA: Continual Orthogonal Low-Rank Adaptation for Class-Incremental Learning
Monu Nagar, Debasis Das
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 soft orthogonal projection matrix P to LoRA (ablation) | COLA · ImageNet-A, ImageNet-R, CUB200 | Acc: +2.6% on ImageNet-A, 2.7% on ImageNet-R, 3.2% on CUB200 | Improved |
| apply LoRA to query/value instead of key/value (ablation) | COLA · ImageNet-R (10 tasks) | Acc (%): 77.13 -> 77.56 | Improved |
| apply LoRA to query/value instead of key/value (ablation) | COLA · ImageNet-A (10 tasks) | Acc (%): 57.86 -> 59.20 | Improved |
| apply LoRA to query/value instead of key/value (ablation) | COLA · CUB200 (20 tasks) | Acc (%): 68.17 -> 69.18 | Improved |
| set residual scaling gamma to 0.20 instead of 0.60 (ablation) | COLA · ImageNet-R (10 tasks) | Acc (%): 73.23 -> 77.56 | Improved |
| set residual scaling gamma to 0.20 instead of 0.08 (ablation) | COLA · ImageNet-A (10 tasks) | Acc (%): 56.55 -> 59.20 | Improved |
| use ViT-B/16 instead of DeiT-S/16 backbone | COLA · ImageNet-R (10 tasks) | Acc (%): 53.92 -> 77.56 | Improved |
| use ViT-B/16 instead of DeiT-B/16 backbone | COLA · CUB200 (10 tasks) | Acc (%): 65.22 -> 74.51 | Improved |
| use COLA instead of SD-LoRA | ImageNet-R (25 tasks) · ImageNet-R (25 tasks) | AAA (%): 74.71 -> 82.11 | Improved |
| use COLA instead of VPT-NSP2 | ImageNet-R (10 tasks) · ImageNet-R (10 tasks) | Acc (%): 78.25 -> 77.56 | Hurt |
| use COLA instead of InfLoRA/SD-LoRA for compute | COLA · Efficiency | GFLOPs: 35.12 -> 32.22 | Improved |
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.
- SD-LoRA — SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025) · better on 8 of 8 shared results
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 1 of 1 shared result
- RanPAC — RanPAC: Random Projections and Pre-trained Models for Continual Learning (NeurIPS 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
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.93)
- CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025 · similarity 0.91)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · similarity 0.91)
- BiLoRA: Almost-Orthogonal Parameter Spaces for Continual Learning (CVPR 2025 · 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)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · 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.
