Research map / Continual learning / Online and replay-based
Continual Learning in Low-rank Orthogonal Subspaces
Arslan Chaudhry, Naeemullah Khan, Puneet Dokania, Philip Torr
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 orthogonal task subspaces without episodic memory | orthog-subspace on Permuted MNIST · Permuted MNIST | Accuracy: 79.5 -> 86.6 | Improved |
| Use orthogonal task subspaces without episodic memory | orthog-subspace on Permuted MNIST · Permuted MNIST | Forgetting: 0.12 -> 0.04 | Improved |
| Use orthogonal task subspaces without episodic memory | orthog-subspace on Rotated MNIST · Rotated MNIST | Accuracy: 70.9 -> 80.1 | Improved |
| Use orthogonal task subspaces without episodic memory | orthog-subspace on Rotated MNIST · Rotated MNIST | Forgetting: 0.24 -> 0.14 | Improved |
| Use orthogonal subspaces, replay memory, and Stiefel-constrained weights | orthog-subspace on Split CIFAR · Split CIFAR | Accuracy: 59.6 -> 64.3 | Improved |
| Use orthogonal subspaces, replay memory, and Stiefel-constrained weights | orthog-subspace on Split miniImageNet · Split miniImageNet | Accuracy: 49.8 -> 51.4 | Improved |
| Add projection to ER-Ring without Stiefel constraint (ablation) | orthog-subspace on Split CIFAR · Split CIFAR | Accuracy: 59.6 -> 61.2 | Improved |
| Add projection to ER-Ring without Stiefel constraint (ablation) | orthog-subspace on Split miniImageNet · Split miniImageNet | Accuracy: 49.8 -> 49.5 | No clear effect |
| Add the Stiefel constraint to projection plus ER (ablation) | orthog-subspace on Split CIFAR · Split CIFAR | Accuracy: 61.2 -> 64.3 | Improved |
| Add the Stiefel constraint to projection plus ER (ablation) | orthog-subspace on Split miniImageNet · Split miniImageNet | Accuracy: 49.5 -> 51.4 | Improved |
| Increase replay memory from 2 to 5 examples per class per task (ablation) | orthog-subspace on Split CIFAR · Split CIFAR | Accuracy: 64.7 -> 67.3 | Improved |
| Increase replay memory from 2 to 5 examples per class per task (ablation) | orthog-subspace on Split miniImageNet · Split miniImageNet | Accuracy: 53.4 -> 58.2 | Improved |
Closest papers by meaning
- BiLoRA: Almost-Orthogonal Parameter Spaces for Continual Learning (CVPR 2025 · similarity 0.92)
- PLAN: Proactive Low-Rank Allocation for Continual Learning (ICCV 2025 · similarity 0.90)
- Continuous Subspace Optimization for Continual Learning (NeurIPS 2025 · similarity 0.90)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.89)
- Continual Learning with Global Alignment (NeurIPS 2024 · similarity 0.89)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · 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.
