Research map / Continual learning / Online and replay-based
Continual Unsupervised Representation Learning
Dushyant Rao, Francesco Visin, Andrei Rusu, Razvan Pascanu, Yee Whye Teh, Raia Hadsell
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 dynamic expansion rather than a fixed mixture size (ablation) | CURL · MNIST sequential | Cluster accuracy: 51% -> 77% | Improved |
| Add mixture generative replay (ablation) | CURL · MNIST sequential | Cluster acc (%): 45.35_{\pm 1.50} -> 77.74_{\pm 1.37} | Improved |
| Add mixture generative replay (ablation) | CURL · Omniglot sequential | Cluster acc (%): 13.36_{\pm 1.06} -> 13.21_{\pm 0.53} | No clear effect |
| Snapshot at 0.1T intervals rather than T intervals (ablation) | CURL with MGR · MNIST sequential | Cluster acc (%): 77.74_{\pm 1.37} -> 49.14_{\pm 3.95} | Hurt |
| Snapshot before dynamic expansion instead of at T intervals (ablation) | CURL with MGR · MNIST sequential | Cluster acc (%): 77.74_{\pm 1.37} -> 57.76_{\pm 1.43} | Hurt |
| Add self-supervised component-consistency loss (ablation) | CURL with MGR · MNIST sequential | Cluster acc (%): 77.74_{\pm 1.37} -> 69.27_{\pm 1.46} | Hurt |
| Add self-supervised component-consistency loss (ablation) | CURL with MGR · Omniglot sequential | Cluster acc (%): 13.21_{\pm 0.53} -> 11.32_{\pm 0.52} | Improved |
| Use continuous drift rather than abrupt sequential class changes (ablation) | CURL with MGR (fixed snapshots) · MNIST | Cluster acc (%): 77.74_{\pm 1.37} -> 79.38_{\pm 4.26} | Improved |
| Use continuous drift rather than abrupt sequential class changes (ablation) | CURL with MGR (dynamic snapshots) · Omniglot | Cluster acc (%): 12.74_{\pm 0.60} -> 13.54_{\pm 0.35} | Improved |
| Increase the expansion threshold, allowing more components (ablation) | CURL · MNIST sequential | Cluster accuracy: 12 to 71 components; up to 84% | Mixed |
| Train CURL on shuffled i.i.d data instead of sequential data | CURL · MNIST | 10-NN error (%): 4.23_{\pm 0.30} vs 4.50_{\pm 0.34} | No clear effect |
Closest papers by meaning
- Representational Continuity for Unsupervised Continual Learning (ICLR 2022 · similarity 0.92)
- Probing Representation Forgetting in Supervised and Unsupervised Continual Learning (CVPR 2022 · similarity 0.91)
- Revisiting Supervision for Continual Representation Learning (ECCV 2024 · similarity 0.89)
- Continual Learning with Global Alignment (NeurIPS 2024 · similarity 0.89)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.87)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.87)
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.
